CN112208573A - Track defect detection system and method based on image recognition - Google Patents
Track defect detection system and method based on image recognition Download PDFInfo
- Publication number
- CN112208573A CN112208573A CN202010982473.1A CN202010982473A CN112208573A CN 112208573 A CN112208573 A CN 112208573A CN 202010982473 A CN202010982473 A CN 202010982473A CN 112208573 A CN112208573 A CN 112208573A
- Authority
- CN
- China
- Prior art keywords
- image
- module
- defect
- rail
- detection
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Granted
Links
Images
Classifications
-
- B—PERFORMING OPERATIONS; TRANSPORTING
- B61—RAILWAYS
- B61K—AUXILIARY EQUIPMENT SPECIALLY ADAPTED FOR RAILWAYS, NOT OTHERWISE PROVIDED FOR
- B61K9/00—Railway vehicle profile gauges; Detecting or indicating overheating of components; Apparatus on locomotives or cars to indicate bad track sections; General design of track recording vehicles
- B61K9/08—Measuring installations for surveying permanent way
- B61K9/10—Measuring installations for surveying permanent way for detecting cracks in rails or welds thereof
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T7/00—Image analysis
- G06T7/0002—Inspection of images, e.g. flaw detection
- G06T7/0004—Industrial image inspection
- G06T7/001—Industrial image inspection using an image reference approach
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T2207/00—Indexing scheme for image analysis or image enhancement
- G06T2207/20—Special algorithmic details
- G06T2207/20081—Training; Learning
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T2207/00—Indexing scheme for image analysis or image enhancement
- G06T2207/20—Special algorithmic details
- G06T2207/20084—Artificial neural networks [ANN]
Landscapes
- Engineering & Computer Science (AREA)
- Mechanical Engineering (AREA)
- Quality & Reliability (AREA)
- Computer Vision & Pattern Recognition (AREA)
- Physics & Mathematics (AREA)
- General Physics & Mathematics (AREA)
- Theoretical Computer Science (AREA)
- Image Analysis (AREA)
- Investigating Materials By The Use Of Optical Means Adapted For Particular Applications (AREA)
- Image Processing (AREA)
Abstract
The invention provides a track defect detection system and method based on image recognition, which are used for solving the problems that the track cannot be comprehensively detected and the detection efficiency is low in the prior art. Firstly, acquiring an image of the surface of a rail under a normal condition on site, classifying image data by using a deep learning algorithm, taking the image data as a standard image, and storing the standard image in a controller; the detection trolley runs on the surface of the track, an encoder generates a pulse signal to be used as a trigger signal to acquire images in real time, the images are preprocessed, then an image recognition is carried out by utilizing a deep learning target detection algorithm to generate an optimal image, the optimal image is compared with a pre-stored standard image, and whether the current track surface has defects or not is judged. The method and the device can quickly and accurately identify whether the surface of the rail has defects, the defect types, the defect positions and the defect time, have accurate detection results, are real-time, efficient, comprehensive and intelligent, and improve the working efficiency of rail detection.
Description
Technical Field
The invention belongs to the field of rail detection, and particularly relates to a rail defect detection system and method based on image recognition.
Background
In transportation systems such as high-speed trains and subways, a track is the basis for train operation, and good track conditions guarantee safe train running. However, in the running process of the train, the contact between the train and the track can cause the abrasion of the track, and the electric spark phenomenon can cause pits and pits on the surface of the track; the vibration phenomenon can cause track cracks and lock catches to be loosened and lost; the problems of rail breakage and large rail gap are caused by the change of temperature difference, the reason of rail expansion and contraction and the technical problem of equipment installation. Therefore, the rail defects are detected, problems are discovered and solved in time, and the safe operation of rail transit is guaranteed.
In the prior art, the detection of the surface defects of the rails is generally only specific to a certain rail, and the detection types of the surface defects are not comprehensive. For the track, the surface defects of the contact rail and the walking rail are mainly detected by manual visual inspection at present, the efficiency is low, the detection precision is not high, the influence of human factors is large, and the simultaneous detection of the surface defects of the contact rail, the walking rail and the lock catch cannot be realized.
Disclosure of Invention
In view of the above-mentioned defects or shortcomings in the prior art, the present invention aims to provide a system and a method for detecting track defects based on image recognition, which can automatically recognize images, generate optimal images of a contact rail, a walking rail and a lock catch at the same time, compare the images with pre-stored normal track standard images, judge whether the track surface has defects, generate optimal images through image acquisition and algorithm, compare the optimal images with images of a main control system, complete the detection of the track surface defects, and realize efficient, accurate and intelligent track detection.
In order to achieve the above purpose, the embodiment of the invention adopts the following technical scheme:
in a first aspect, an embodiment of the present invention provides a method for detecting a track defect based on image recognition, where the method for detecting a track defect includes the following steps:
step S1, acquiring image data on site, wherein the image data comprises images of the left and right contact rails of the track, the walking rails and the surfaces of the lock catches under normal working conditions, as well as images of track defects, track abrasion, lock catch defects and lock catch loss, and meanwhile, intensifying the images; classifying the data, labeling different detection points, iterating field data by using a deep learning algorithm, and training a model; as a standard image, stored in the controller;
step S2, detecting that the trolley runs on the surface of the track, generating a pulse signal by an image acquisition module through an encoder, acquiring images by taking the pulse signal as a trigger signal, acquiring real-time images and position information of the surfaces of the contact rail, the walking rail and the lock catch in real time, and uploading the images and the position information to a controller;
step S3, preprocessing the real-time image;
step S4, carrying out image recognition on the preprocessed real-time image by using a deep learning target detection algorithm to generate an optimal image;
step S5, comparing the optimal image with a pre-stored standard image, judging whether the current track surface has defects according to the comparison result, and returning to the step S2 when no defects exist; when there is a defect, proceed to step S6;
and step S6, confirming the defect type, reporting the defect type information and the defect image to the controller, sending an alarm signal and storing alarm information by the controller, and displaying the real defect image and the position information on a system interface.
In the scheme, the image acquisition module generates a pulse signal through the encoder, when the detection trolley pushes forwards, the encoder sends a signal once when the wheels rotate for one circle, and the encoder triggers the image acquisition module to generate a signal once.
In the scheme, the image acquisition is completed by a left contact rail image acquisition submodule, a right contact rail image acquisition submodule, a left walking rail image acquisition submodule, a right walking rail image acquisition submodule and a lock catch image acquisition submodule which are detachably arranged on the detection trolley.
In the scheme, the random image acquisition submodule consists of a camera, a lens, a light source and a closed shell.
In the above scheme, the pretreatment comprises: and performing geometric transformation, noise removal and image enhancement on the image.
In the scheme, the noise is removed, the median filtering algorithm is used for performing noise reduction on the acquired image, the clear outline of the image is kept, the second-order difference method is used for performing contrast enhancement processing on the image, and the expansion and corrosion operators in mathematical morphology are used for processing the noise and the cavity in the image.
In the foregoing solution, the YOLOV3 is selected for the deep learning in step S4 to perform the target detection, and the image recognition is performed by using a deep learning algorithm, using a residual error network in combination with an FPN network structure, aggregating feature maps of corresponding sizes at an early stage of the network after the two feature maps are sampled after the two feature maps are networked, and obtaining a prediction result after the aggregation is performed through a convolutional network, thereby generating an image recognition model; the model omits the extraction branch of the candidate frame, directly completes the feature extraction, the regression and the classification of the candidate frame in the same non-branched convolution network, and outputs the optimal image.
In a second aspect, an embodiment of the present invention further provides an image recognition-based rail defect detection system, where the detection system includes: the detection trolley comprises a detection trolley, a controller, an encoder and an image acquisition module, wherein the image acquisition module consists of a left contact rail image acquisition sub-module, a right contact rail image acquisition sub-module, a left walking rail image acquisition sub-module, a right walking rail image acquisition sub-module and a lock catch image acquisition sub-module; the controller, the encoder and the image acquisition module are assembled on the detection trolley, and the detection trolley runs on the current track; wherein the content of the first and second substances,
the controller comprises a control center, a memory, an image processing module, an image comparison module and a defect alarm module; the control center is simultaneously connected with the memory, the image processing module, the image comparison module and the defect alarm module, the image processing module is connected with the image comparison module, and the image comparison module is connected with the defect alarm module and the control center of the memory;
the control center is connected with the image acquisition submodule, and the image acquisition module is connected with the encoder; the image acquisition modules are connected with the memory and the image comparison module;
the image acquisition module is used for acquiring images of the surfaces of the left and right contact rails, the walking rails and the latches of the subway under normal working conditions, and acquiring a rail defect image, a rail abrasion image, a latch defect image and a latch loss image as standard images and storing the standard images in the controller; the controller is used for acquiring real-time images of the surfaces of the contact rail, the walking rail and the lock catch in real time after receiving a pulse signal of the encoder, and uploading the real-time images to the controller;
the encoder is used for generating pulse signals under the control of the rotation of wheels of the rail detection trolley and starting the image acquisition module;
the image processing module is used for preprocessing the real-time image acquired by the image acquisition module, recognizing the preprocessed real-time image by using a deep learning target detection algorithm, generating an optimal image and sending the optimal image to the image comparison module;
the image comparison module is used for calling the standard image in the memory after receiving the optimal image, comparing the optimal image with the standard image and judging whether the current rail surface has defects or not according to the comparison result; when no defect exists, feeding back non-defective information to the control center; when the defect exists, the comparison result is sent to a defect alarm module, and the defect, the defect type and the position information are fed back to the control center;
and the defect alarm module is used for classifying the defects after receiving the comparison result of the image comparison module, starting alarm according to the defect type, recording real images and positions of the surface in real time, and feeding back the real images and the positions to the image comparison module.
In the above scheme, the controller, the encoder and the image acquisition module are detachably mounted on the detection trolley.
In the above scheme, the deep learning target detection algorithm of the image processing module selects YOLOV3 for target detection.
The invention has the following beneficial effects:
the track defect detection method based on image recognition of the embodiment of the invention simultaneously carries out online detection on the surface defects of the left contact rail, the right contact rail, the walking rail and the lock catch, and quickly and accurately recognizes whether the surface of the track has defects, defect types and defect positions; an intelligent image recognition technology and a big data processing technology are integrated, so that the working difficulty of detection personnel is reduced, and the accuracy of detection data is improved; by adopting the traditional algorithm and the deep learning target detection algorithm, the image acquisition frequency is high, the defect judgment precision is high, the real-time performance is strong, and the working efficiency is high.
Drawings
Other features, objects and advantages of the invention will become more apparent upon reading of the detailed description of non-limiting embodiments made with reference to the following drawings:
FIG. 1 is a flowchart of a track defect detection method based on image recognition according to an embodiment of the present invention;
FIG. 2 is a schematic structural diagram of a track defect detection system based on image recognition according to an embodiment of the present invention;
FIG. 3 is an assembly view of a detection carriage in the rail defect detection system of FIG. 2.
Detailed Description
The present invention will be described in further detail with reference to the accompanying drawings and examples. It is to be understood that the specific embodiments described herein are merely illustrative of the relevant invention and not restrictive of the invention. It should be noted that, for convenience of description, only the portions related to the present invention are shown in the drawings.
It should be noted that the embodiments and features of the embodiments may be combined with each other without conflict. The present invention will be described in detail below with reference to the embodiments with reference to the attached drawings.
Fig. 1 shows a flowchart of a track defect detection method based on image recognition according to an embodiment of the present invention. As shown in fig. 1, the defect detection method includes the following steps:
step S1, acquiring image data on site, wherein the image data comprises images of the left and right contact rails of the track, the walking rails and the surfaces of the lock catches under normal working conditions, as well as images of track defects, track abrasion, lock catch defects and lock catch loss, and meanwhile, intensifying the images; classifying the image data acquired on site, labeling different detection points, iterating the image data acquired on site by using a deep learning algorithm, and training a model; the method comprises the steps of taking an image collected on site as a standard image, storing the standard image in a controller, and calling the standard image from the controller every time of inspection later.
The image target features are identified through a deep learning algorithm, an end-to-end framework is used for designing a defect and lock catch target identification model, a newly designed residual error Network is adopted and combined with a Feature Pyramid Network (FPN) Network structure, after two Feature maps after the Network are sampled, Feature maps with corresponding sizes in the early stage of the Network are aggregated and then are subjected to convolution Network, and a prediction result is obtained. These improvements allow the model to be made in less time and with greater accuracy.
The deep learning algorithm trains the model, the convolutional network is essentially an input-to-output mapping, and can learn a large number of input-to-output mapping relationships without any precise mathematical expressions between the inputs and the outputs, and the network has the capability of mapping between input-output pairs as long as the convolutional network is trained with a known pattern.
The convolutional network performs supervised training so that its sample set is formed of pairs of input vectors and ideal output vectors. All these vector pairs should be the actual "running" results from the system that the network is about to simulate. They are collected from the actual operating system.
Before training is started, all weights should be initialized with some different small random number. The small random number is used for ensuring that the network does not enter a saturation state due to overlarge weight value, so that training fails; "different" is used to ensure that the network can learn normally.
And step S2, formally starting the inspection, acquiring real-time images, detecting the running of the trolley on the surface of the track, generating pulse signals by the image acquisition module through the encoder, triggering the image acquisition module to acquire images by using the pulse signals as trigger signals, acquiring real-time images and position information of the surfaces of the left contact rail, the right contact rail, the walking rail and the lock catch in real time, and uploading the real-time images and the position information to the controller.
Preferably, the detection trolley is pushed forwards, the encoder sends out a pulse signal every time the wheels rotate for one circle, and the image acquisition system photographs the surface of the track according to the pulse signal provided by the encoder to acquire a real image of the surface of the track.
In this step, image acquisition is performed by the image acquisition module. The image acquisition module adopts a modular design and comprises a left contact rail acquisition sub-module, a right contact rail acquisition sub-module, a walking rail acquisition sub-module and a lock catch acquisition sub-module. The four modules are arranged at corresponding positions of the detection trolley, and the trolley finishes respective acquisition work in the process of advancing on the track. Meanwhile, all modules are not interfered with each other, are flexibly arranged according to the detection requirement of the surface defects of the track, and are easy to disassemble, assemble, troubleshoot and maintain.
Each image acquisition submodule consists of a camera, a lens, a light source and a closed shell. The image acquisition submodule adopts a closed shell design, and is moisture-proof, dust-proof and high in environmental adaptability. The design of a visible light camera and a light source structure is adopted, the information of the surface defects (pits, falling, cracks and abrasion) of the contact rail and the traveling rail is recorded in the whole process, and the information of the surface defects (cracks, loss and bolt looseness) of the lock catch is recorded in the whole process.
And step S3, preprocessing the real-time image.
In this step, the pretreatment comprises: and performing geometric transformation, noise removal and image enhancement on the image.
Preferably, the median filtering algorithm is used for carrying out noise reduction on the acquired image, meanwhile, the clear outline of the image is kept, the second-order difference method is used for carrying out contrast enhancement processing on the image, and the mathematical morphological expansion and corrosion operators are used for processing small noise and holes in the image.
And step S4, performing image recognition on the preprocessed real-time image by using a deep learning target detection algorithm to generate an optimal image.
Preferably, in this step, YOLOV3 is selected for object detection in deep learning, and an optimal image is generated.
Step S41 is to perform image recognition on the real-time image to generate an image recognition model.
The target detection algorithm for deep learning uses a fully newly designed residual error network and combines with an FPN network structure, two feature maps after the network are sampled are aggregated in the feature maps with corresponding sizes in the early stage of the network, and a prediction result is obtained after the aggregation is performed through a convolution network, so that an image recognition model is generated.
And step S42, outputting an optimal image, omitting a candidate frame extraction branch from the generated image recognition model, and directly completing feature extraction, candidate frame regression and classification in the same non-branch convolution network, so that the network structure becomes simple, the detection speed is greatly improved, and the optimal image is output.
Step S5, comparing the optimal image with a pre-stored standard image, judging whether the current track surface has defects according to the comparison result, and returning to the step S2 when no defects exist; when there is a defect, the process proceeds to step S6.
And step S6, confirming the defect type, reporting the defect type information and the defect image to the controller, sending an alarm signal and storing alarm information by the controller, and displaying the real defect image and the position information on a system interface.
According to the technical scheme, the image is automatically identified, the defective image is transmitted to the main control system, the current picture can be stored and transmitted to the controller under the condition that the controller is triggered, the track surface defect image and the report are obtained, the contact track surface defect, the walking track surface defect, the fastener loss and the breakage are detected, and the detection process is efficient, accurate and intelligent.
Fig. 2 shows a structure of a track defect detection system based on image recognition provided by an embodiment of the invention. As shown in fig. 2, the track defect detecting system based on image recognition comprises: the detection trolley comprises a detection trolley, a controller, an encoder and an image acquisition module, wherein the image acquisition module consists of a left contact rail image acquisition submodule, a right contact rail image acquisition submodule, a left walking rail image acquisition submodule, a right walking rail image acquisition submodule and a lock catch image acquisition submodule, and the controller comprises a control center, a memory, an image processing module, an image comparison module and a defect alarm module. The track defect detection system based on image recognition of the present embodiment is used to realize the detection method in the above embodiments.
As shown in fig. 3, the controller, the encoder and the image acquisition module are detachably mounted on the detection trolley, and the detection trolley is adaptive to the current track to be detected and runs on the current track.
In the controller, a control center is simultaneously connected with a memory, an image processing module, an image comparison module and a defect alarm module, the image processing module is connected with the image comparison module, the image comparison module is connected with the defect alarm module and the control center of the memory, and data in the memory is called through the control center. The control center of the controller is connected with an encoder, and the encoder is connected with an image acquisition module; the image acquisition modules are connected with the memory and the image comparison module.
In the initialization stage of the detection trolley, the image acquisition module acquires images of the surfaces of the left and right contact rails, the walking rails and the lock catches of the subway under normal working conditions, the images serve as standard images, and the standard images are stored in a memory of the controller.
The control center of the controller selects image acquisition frequency according to the running condition of the detection trolley and the feedback of the image comparison module, controls the encoder to generate pulse signals, and the image acquisition submodule connected with the encoder acquires images according to the pulse signal frequency and stores the acquired images in a memory of the controller in real time.
And the image processing module is used for preprocessing the real-time image after receiving the real-time image, recognizing the preprocessed real-time image by using a deep learning target detection algorithm, generating an optimal image and sending the optimal image to the image comparison module. The pretreatment comprises the following steps: and performing geometric transformation, noise removal and image enhancement on the image. Preferably, the image processing module comprises a preprocessing sub-module and an optimal image generation sub-module. Preferably, the deep learning selects YOLOV3 for target detection, and generates an optimal image.
After receiving the optimal image, the image comparison module calls a standard image in the memory, compares the optimal image with the standard image, and judges whether the current rail surface has defects according to the comparison result; when no defect exists, feeding back non-defective information to the control center; and when the defect exists, the comparison result is sent to a defect alarm module, and the information of the defect, the defect type and the position is fed back to the control center.
And the defect alarm module classifies the defects after receiving the comparison result of the image comparison module, starts alarm according to the defect type, records real images and positions of the surface in real time and feeds back the real images and the positions to the image comparison module.
As described above, the track defect detecting system based on image recognition may further include a display device. The display device is connected with the control center. When the control center receives defect information fed back by the image comparison module, the defect information is displayed, and the defect information at least comprises: current track image, standard image, defect type, location, and time.
The track defect detection system based on image recognition provided by the embodiment has the advantages that the device is detachable, can be flexibly distributed in a self-adaptive way according to the track surface defect detection requirement, and is easy to disassemble, assemble, troubleshoot and maintain; the detection process is reasonable, the detection result is accurate, real-time, efficient, comprehensive and intelligent, and the working efficiency of track detection is improved.
The foregoing description is only exemplary of the preferred embodiments of the invention and is illustrative of the principles of the technology employed. It will be appreciated by those skilled in the art that the scope of the invention herein disclosed is not limited to the particular combination of features described above, but also encompasses other arrangements formed by any combination of the above features or their equivalents without departing from the spirit of the invention. For example, the above features and (but not limited to) features having similar functions disclosed in the present invention are mutually replaced to form the technical solution.
Claims (10)
1. A track defect detection method based on image recognition is characterized by comprising the following steps:
step S1, acquiring image data on site, wherein the image data comprises images of the left and right contact rails of the track, the walking rails and the surfaces of the lock catches under normal working conditions, as well as images of track defects, track abrasion, lock catch defects and lock catch loss, and meanwhile, intensifying the images; classifying the image data acquired on site, labeling different detection points, iterating the image data acquired on site by using a deep learning algorithm, and training a model; taking the image collected on site as a standard image and storing the standard image in a controller;
step S2, detecting that the trolley runs on the surface of the track, generating a pulse signal by the image acquisition module through the encoder, acquiring images by taking the pulse signal as a trigger signal, acquiring real-time images and position information of the surfaces of the contact rail, the walking rail and the lock catch in real time, and uploading the images and the position information to the controller;
step S3, preprocessing the real-time image;
step S4, carrying out image recognition on the preprocessed real-time image by using a deep learning target detection algorithm to generate an optimal image;
step S5, comparing the optimal image with a pre-stored standard image, judging whether the current track surface has defects according to the comparison result, and returning to the step S2 when no defects exist; when there is a defect, proceed to step S6;
and step S6, confirming the defect type, reporting the defect type information and the defect image to the controller, sending an alarm signal and storing alarm information by the controller, and displaying the real defect image and the position information on a system interface.
2. The method as claimed in claim 1, wherein the image capturing module generates a pulse signal through an encoder, and the encoder generates a signal once every time the detection trolley is pushed forward and the wheel rotates once, and the encoder triggers the image capturing module to generate a signal once.
3. The image recognition-based rail defect detection method according to claim 1, wherein the image acquisition is performed by a left contact rail image acquisition sub-module, a right contact rail image acquisition sub-module, a left running rail image acquisition sub-module, a right running rail image acquisition sub-module and a locking buckle image acquisition sub-module which are detachably mounted on the detection trolley.
4. The rail defect detection method based on image recognition is characterized in that any image acquisition submodule consists of a camera, a lens, a light source and a closed shell.
5. The image recognition-based rail defect detection method according to claim 1, wherein the preprocessing comprises: and performing geometric transformation, noise removal and image enhancement on the image.
6. The method of claim 5, wherein the removing noise comprises performing noise reduction on the collected image by using a median filtering algorithm while keeping a clear outline of the image, performing contrast enhancement on the image by using a second-order difference method, and processing noise and holes in the image by using mathematical morphological dilation and erosion operators.
7. The image recognition-based track defect detection method of claim 1,
in the step S4, the deep learning target detection algorithm selects YOLOV3 for target detection;
the image recognition is to use a deep learning algorithm, use a residual error network and combine a feature pyramid network FPN network structure, aggregate feature maps with corresponding sizes in the early stage of the network after two feature maps are sampled after the network, and obtain a prediction result after the aggregation and the convolution network, so as to generate an image recognition model; the image recognition model omits a candidate frame extraction branch, directly completes feature extraction, candidate frame regression and classification in the same non-branch convolution network, and outputs an optimal image.
8. An image recognition-based rail defect detection system, the detection system comprising: the detection trolley comprises a detection trolley, a controller, an encoder and an image acquisition module, wherein the image acquisition module consists of a left contact rail image acquisition sub-module, a right contact rail image acquisition sub-module, a left walking rail image acquisition sub-module, a right walking rail image acquisition sub-module and a lock catch image acquisition sub-module; the controller, the encoder and the image acquisition module are assembled on the detection trolley, and the detection trolley runs on the current track; wherein the content of the first and second substances,
the controller comprises a control center, a memory, an image processing module, an image comparison module and a defect alarm module; the control center is simultaneously connected with the memory, the image processing module, the image comparison module and the defect alarm module, the image processing module is connected with the image comparison module, and the image comparison module is connected with the defect alarm module and the control center of the memory;
the control center is connected with the image acquisition submodule, and the image acquisition module is connected with the encoder; the image acquisition modules are connected with the memory and the image comparison module;
the image acquisition module is used for acquiring images of the surfaces of the left and right contact rails, the walking rails and the latches of the subway under normal working conditions, and acquiring a rail defect image, a rail abrasion image, a latch defect image and a latch loss image as standard images and storing the standard images in the controller; the controller is used for acquiring real-time images of the surfaces of the contact rail, the walking rail and the lock catch in real time after receiving a pulse signal of the encoder, and uploading the real-time images to the controller;
the encoder is used for generating pulse signals under the control of the rotation of wheels of the rail detection trolley and starting the image acquisition module;
the image processing module is used for preprocessing the real-time image acquired by the image acquisition module, recognizing the preprocessed real-time image by using a deep learning target detection algorithm, generating an optimal image and sending the optimal image to the image comparison module;
the image comparison module is used for calling the standard image in the memory after receiving the optimal image, comparing the optimal image with the standard image and judging whether the current rail surface has defects or not according to the comparison result; when no defect exists, feeding back non-defective information to the control center; when the defect exists, the comparison result is sent to a defect alarm module, and the defect, the defect type and the position information are fed back to the control center;
and the defect alarm module is used for classifying the defects after receiving the comparison result of the image comparison module, starting alarm according to the defect type, recording real images and positions of the surface in real time, and feeding back the real images and the positions to the image comparison module.
9. The image recognition-based rail defect detection system of claim 8, wherein the controller, encoder and image acquisition module are removably mounted on the inspection trolley.
10. The image recognition-based rail defect detection system of claim 8, wherein the deep learning target detection algorithm of the image processing module selects YOLOV3 for target detection.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202010982473.1A CN112208573B (en) | 2020-09-17 | 2020-09-17 | Track defect detection system and method based on image recognition |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202010982473.1A CN112208573B (en) | 2020-09-17 | 2020-09-17 | Track defect detection system and method based on image recognition |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN112208573A true CN112208573A (en) | 2021-01-12 |
| CN112208573B CN112208573B (en) | 2022-04-01 |
Family
ID=74050426
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202010982473.1A Active CN112208573B (en) | 2020-09-17 | 2020-09-17 | Track defect detection system and method based on image recognition |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN112208573B (en) |
Cited By (8)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN110901474A (en) * | 2019-11-07 | 2020-03-24 | 毛军光 | Railway cable support |
| CN113112501A (en) * | 2021-05-11 | 2021-07-13 | 上海市东方海事工程技术有限公司 | Vehicle-mounted track inspection device and method based on deep learning |
| CN113776461A (en) * | 2021-09-09 | 2021-12-10 | 北京京东乾石科技有限公司 | Three-dimensional detection equipment for detecting surface of track |
| CN114308721A (en) * | 2021-12-17 | 2022-04-12 | 深圳市杰普特光电股份有限公司 | Optical detection method, system and storage medium |
| CN114381976A (en) * | 2022-01-20 | 2022-04-22 | 武汉大学 | Expert system-based self-adaptive water jet steel rail grinding method and equipment |
| CN114475369A (en) * | 2021-12-30 | 2022-05-13 | 威海南海数字产业研究院有限公司 | Urban rail transit contact rail system with noise reduction function |
| CN115100110A (en) * | 2022-05-20 | 2022-09-23 | 厦门微亚智能科技有限公司 | Defect detection method, device and equipment for polarized lens and readable storage medium |
| CN115797914A (en) * | 2023-02-02 | 2023-03-14 | 武汉科技大学 | Metallurgical crane trolley track surface defect detection system |
Citations (4)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN102897192A (en) * | 2012-10-18 | 2013-01-30 | 成都唐源电气有限责任公司 | Detection system for urban railway traffic contact rail and detection method thereof |
| CN110044645A (en) * | 2019-05-07 | 2019-07-23 | 中国铁道科学研究院集团有限公司 | Protective cover board for contact rails condition detecting system and method |
| CN111366082A (en) * | 2020-04-16 | 2020-07-03 | 青岛地铁集团有限公司运营分公司 | Movable contact rail detection device and application method thereof |
| CN111402209A (en) * | 2020-03-03 | 2020-07-10 | 广州中国科学院先进技术研究所 | U-Net-based high-speed railway steel rail damage detection method |
-
2020
- 2020-09-17 CN CN202010982473.1A patent/CN112208573B/en active Active
Patent Citations (4)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN102897192A (en) * | 2012-10-18 | 2013-01-30 | 成都唐源电气有限责任公司 | Detection system for urban railway traffic contact rail and detection method thereof |
| CN110044645A (en) * | 2019-05-07 | 2019-07-23 | 中国铁道科学研究院集团有限公司 | Protective cover board for contact rails condition detecting system and method |
| CN111402209A (en) * | 2020-03-03 | 2020-07-10 | 广州中国科学院先进技术研究所 | U-Net-based high-speed railway steel rail damage detection method |
| CN111366082A (en) * | 2020-04-16 | 2020-07-03 | 青岛地铁集团有限公司运营分公司 | Movable contact rail detection device and application method thereof |
Non-Patent Citations (2)
| Title |
|---|
| 孙明华等: "基于深度可分离卷积的地铁隧道巡检视频分析", 《计算机工程与科学》 * |
| 赵永强等: "深度学习目标检测方法综述", 《中国图象图形学报》 * |
Cited By (10)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN110901474A (en) * | 2019-11-07 | 2020-03-24 | 毛军光 | Railway cable support |
| CN113112501A (en) * | 2021-05-11 | 2021-07-13 | 上海市东方海事工程技术有限公司 | Vehicle-mounted track inspection device and method based on deep learning |
| CN113112501B (en) * | 2021-05-11 | 2023-01-20 | 上海市东方海事工程技术有限公司 | Vehicle-mounted track inspection device and method based on deep learning |
| CN113776461A (en) * | 2021-09-09 | 2021-12-10 | 北京京东乾石科技有限公司 | Three-dimensional detection equipment for detecting surface of track |
| CN114308721A (en) * | 2021-12-17 | 2022-04-12 | 深圳市杰普特光电股份有限公司 | Optical detection method, system and storage medium |
| CN114308721B (en) * | 2021-12-17 | 2024-01-26 | 深圳市杰普特光电股份有限公司 | Optical detection method, system and storage medium |
| CN114475369A (en) * | 2021-12-30 | 2022-05-13 | 威海南海数字产业研究院有限公司 | Urban rail transit contact rail system with noise reduction function |
| CN114381976A (en) * | 2022-01-20 | 2022-04-22 | 武汉大学 | Expert system-based self-adaptive water jet steel rail grinding method and equipment |
| CN115100110A (en) * | 2022-05-20 | 2022-09-23 | 厦门微亚智能科技有限公司 | Defect detection method, device and equipment for polarized lens and readable storage medium |
| CN115797914A (en) * | 2023-02-02 | 2023-03-14 | 武汉科技大学 | Metallurgical crane trolley track surface defect detection system |
Also Published As
| Publication number | Publication date |
|---|---|
| CN112208573B (en) | 2022-04-01 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN112208573B (en) | Track defect detection system and method based on image recognition | |
| Yanan et al. | Rail surface defect detection method based on YOLOv3 deep learning networks | |
| Karakose et al. | A new computer vision based method for rail track detection and fault diagnosis in railways | |
| US9050984B2 (en) | Anomalous railway component detection | |
| EP2697738B1 (en) | Method and system of rail component detection using vision technology | |
| Rizvi et al. | Crack detection in railway track using image processing | |
| CN105260744B (en) | The automatic on-line diagnostic method and system of a kind of goods train coupler yoke key position failure | |
| CN111080598B (en) | Bolt and nut missing detection method for coupler yoke key safety crane | |
| CN112528861A (en) | Foreign matter detection method and device applied to track bed in railway tunnel | |
| Lu et al. | Automatic fault detection of multiple targets in railway maintenance based on time-scale normalization | |
| Chen et al. | A hybrid deep learning based framework for component defect detection of moving trains | |
| Zhang et al. | Real-time vision-based system of fault detection for freight trains | |
| CN114565845A (en) | Intelligent inspection system for underground tunnel | |
| CN114998244A (en) | Intelligent track beam finger-shaped plate inspection system and method based on computer vision | |
| CN113788051A (en) | Train on-station running state monitoring and analyzing system | |
| CN112419289A (en) | Intelligent detection method for urban subway rail fastener defects | |
| Wu et al. | Automatic railroad track components inspection using hybrid deep learning framework | |
| CN115527170A (en) | Method and system for identifying closing fault of door stopper handle of automatic freight car derailing brake device | |
| Mumbelli et al. | An application of Generative Adversarial Networks to improve automatic inspection in automotive manufacturing | |
| Di Summa et al. | A review on deep learning techniques for railway infrastructure monitoring | |
| CN116456075A (en) | Automatic inspection system for monitoring video quality | |
| CN113012113B (en) | Automatic detection method for bolt looseness of high-speed rail contact network power supply equipment | |
| Iftikhar et al. | Unsupervised detection of rail surface defects and rail-head anomalies using Auto-encoder | |
| JP7470864B2 (en) | Train inspection system and method | |
| JP2019142304A (en) | Fallen object detection device and fallen object detection method |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| PB01 | Publication | ||
| PB01 | Publication | ||
| SE01 | Entry into force of request for substantive examination | ||
| SE01 | Entry into force of request for substantive examination | ||
| GR01 | Patent grant | ||
| GR01 | Patent grant |