CN111539421A - Deep learning-based railway locomotive number identification method - Google Patents
Deep learning-based railway locomotive number identification method Download PDFInfo
- Publication number
- CN111539421A CN111539421A CN202010260371.9A CN202010260371A CN111539421A CN 111539421 A CN111539421 A CN 111539421A CN 202010260371 A CN202010260371 A CN 202010260371A CN 111539421 A CN111539421 A CN 111539421A
- Authority
- CN
- China
- Prior art keywords
- car number
- car
- deep learning
- network
- picture
- 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.)
- Pending
Links
- 238000013135 deep learning Methods 0.000 title claims abstract description 31
- 238000000034 method Methods 0.000 title claims abstract description 26
- 230000003137 locomotive effect Effects 0.000 title claims abstract description 24
- 238000001514 detection method Methods 0.000 claims abstract description 33
- 238000000605 extraction Methods 0.000 claims description 5
- 230000009466 transformation Effects 0.000 claims description 4
- 239000003086 colorant Substances 0.000 abstract description 5
- 238000010586 diagram Methods 0.000 description 1
- 238000005286 illumination Methods 0.000 description 1
- 238000007689 inspection Methods 0.000 description 1
- 238000011835 investigation Methods 0.000 description 1
- 238000003672 processing method Methods 0.000 description 1
- 238000013024 troubleshooting Methods 0.000 description 1
Images
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
- G06V20/00—Scenes; Scene-specific elements
- G06V20/60—Type of objects
- G06V20/62—Text, e.g. of license plates, overlay texts or captions on TV images
- G06V20/63—Scene text, e.g. street names
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F18/00—Pattern recognition
- G06F18/20—Analysing
- G06F18/24—Classification techniques
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T5/00—Image enhancement or restoration
- G06T5/80—Geometric correction
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
- G06V10/00—Arrangements for image or video recognition or understanding
- G06V10/40—Extraction of image or video features
- G06V10/46—Descriptors for shape, contour or point-related descriptors, e.g. scale invariant feature transform [SIFT] or bags of words [BoW]; Salient regional features
- G06V10/462—Salient features, e.g. scale invariant feature transforms [SIFT]
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
- G06V30/00—Character recognition; Recognising digital ink; Document-oriented image-based pattern recognition
- G06V30/10—Character recognition
Landscapes
- Engineering & Computer Science (AREA)
- Theoretical Computer Science (AREA)
- Physics & Mathematics (AREA)
- General Physics & Mathematics (AREA)
- Data Mining & Analysis (AREA)
- Multimedia (AREA)
- Computer Vision & Pattern Recognition (AREA)
- Bioinformatics & Cheminformatics (AREA)
- Evolutionary Computation (AREA)
- Evolutionary Biology (AREA)
- General Engineering & Computer Science (AREA)
- Bioinformatics & Computational Biology (AREA)
- Artificial Intelligence (AREA)
- Life Sciences & Earth Sciences (AREA)
- Image Processing (AREA)
- Image Analysis (AREA)
Abstract
A railway locomotive number identification method based on deep learning relates to the technical field of railway vehicle fault identification. The invention aims to solve the problems that in the process of manually identifying the locomotive number, the detection result is low in precision and efficiency, and the timely searching of the fault reason is not facilitated. According to the method for identifying the railway locomotive number based on deep learning, the small characters in the locomotive number can be effectively identified by adopting a mode of firstly carrying out rough positioning on the target detection network and then carrying out identification in the positioned screenshot, so that the identification accuracy is improved, and missing detection and wrong detection are avoided. Due to the fact that the types of the colors of the vehicles are too many, the contrast of the car numbers of the colors shot by the black-and-white camera is not high. The invention adopts the color area-array cameras on the two sides of the vehicle body to shoot the vehicle number image, and can effectively avoid the influence of strong sunlight interference, rain, snow and other severe weather on the single-side camera.
Description
Technical Field
The invention belongs to the technical field of railway vehicle fault identification.
Background
The train number of a railway locomotive is an important identification feature of the locomotive, is accompanied with the complete operation process of a truck, and plays a very critical role in the operation process of the railway.
At present, the fault detection of the railway freight car generally adopts a manual troubleshooting mode. The investigation process is greatly influenced by factors such as the business quality, the responsibility and the labor intensity of operators, so that the conditions of missing inspection, operation simplification and the like are easy to occur.
Disclosure of Invention
The invention provides a railway locomotive number identification method based on deep learning, aiming at solving the problems that in the process of manually identifying the railway locomotive number, the detection result is low in precision and efficiency, and the failure reason is not easy to find in time.
The method for identifying the locomotive number based on deep learning comprises the following steps:
the method comprises the following steps: a training set is established, and the training set is established,
collecting car number pictures of different types of vehicles at two sides of the vehicle at different time and under different environments, establishing a sample library by utilizing all the car number pictures,
respectively marking the position of the car number in each car number picture, generating corresponding label files, taking all the car number pictures and the corresponding label files as a training set of the preposed target detection network,
intercepting the car number picture according to the position of the car number in each car number picture to obtain a screenshot containing the car number, marking characters of the car number in the screenshot, marking different characters as different types, and taking all the screenshots and the types corresponding to the characters in the screenshots as a training set of a car number recognition network;
step two: the training of the weight is carried out,
training the target detection network using data in the training set of the pre-target detection network,
training a car number recognition deep learning network by using data in a training set of the car number recognition network;
step three: the collection of the image is carried out,
collecting car number images on two sides of a vehicle to be detected, and adjusting the two car number images into a picture to be detected with pixels of 512 multiplied by 512;
step four: the identification of the number of the car,
inputting two pictures to be detected into a target detection network to obtain a position frame of the car number in the pictures to be detected,
intercepting the detection picture according to the position frame of the car number in the picture to be detected to obtain two car number screenshots, respectively inputting the two car number screenshots into a car number recognition deep learning network to obtain car number types corresponding to more car number screenshots,
step five: as a result of the determination,
judging whether the confidence degrees of the two car number types are the same,
if so, the two screenshots of the car numbers have the same car number type and can be used as the output result of the car numbers,
otherwise, the type of the car number with high confidence coefficient is used as a car number output result.
After the car number images at the two sides of the vehicle to be detected are collected, firstly, the two car number images are respectively subjected to projection transformation to correct the image distortion,
then, the color, brightness and contrast of the two car number images are respectively adjusted, the color of the images is corrected,
and finally, adjusting image pixels to obtain the picture to be detected.
And when the target detection network outputs the position frames where the plurality of car numbers are located, taking the position frame with the maximum reliability as a final result.
In practical application, because some characters in the car number are too small in the image, the accuracy of the small target recognition detection in the deep learning field is not high all the time. Therefore, the method for identifying the railway locomotive number based on deep learning, provided by the invention, has the advantages that the small characters in the locomotive number can be effectively identified by adopting a mode of firstly carrying out rough positioning on the target detection network and then carrying out identification in the positioned screenshot, the identification accuracy is improved, and missing detection and wrong detection are avoided.
Due to the fact that the types of the colors of the vehicles are too many, the contrast of the car numbers of the colors shot by the black-and-white camera is not high. The invention adopts the color area-array cameras on the two sides of the vehicle body to shoot the vehicle number image, and can effectively avoid the influence of strong sunlight interference, rain, snow and other severe weather on the single-side camera.
Drawings
FIG. 1 is a network training flow diagram;
fig. 2 is a flow chart of car number identification.
Detailed Description
As the types of the railway locomotives are various, the writing fonts, the sizes, the colors, the positions and the like of the locomotive numbers of different types are different. The conventional image processing method has difficulty in recognizing all kinds of car numbers.
With the great improvement of the processing performance of chip hardware, a foundation is provided for the complex computation of a deep network. The deep learning is widely applied to the field of image processing, and compared with the traditional mode, the deep learning integrates the feature learning into the process of establishing the model, so that the accuracy and efficiency of fault detection can be effectively improved.
Therefore, the invention provides the following embodiment, and the car number is identified by adopting a deep learning mode.
The first embodiment is as follows: specifically describing the present embodiment with reference to fig. 1 and 2, the method for identifying a railroad locomotive number based on deep learning according to the present embodiment includes the steps of:
the method comprises the following steps: a training set is established, and the training set is established,
the high-definition image acquisition equipment arranged on the two sides of the track is used for acquiring a large number of car number pictures of vehicles with different types on the two sides of the vehicles at different time and in different environments, and the car number pictures are used as initial data sets. In order to enrich the types and the quantity of samples, the pictures in the initial data set are respectively subjected to transformation such as stretching and rotation, so that the initial data set is amplified, and the robustness and the adaptability of the training result are improved. And finally, establishing a sample library by using all car number pictures.
And respectively marking the position of the car number in each car number picture, generating corresponding label files, and taking all the car number pictures and the corresponding label files as a training set of the preposed target detection network.
And intercepting the car number picture according to the position of the car number in each car number picture to obtain a screenshot containing the car number, marking the characters of the car number in the screenshot, marking different characters as different types, and taking all the screenshots and the types corresponding to the characters in the screenshots as a training set of a car number recognition network.
Step two: the training of the weight is carried out,
the target detection network for positioning the train number position adopts an SSD deep learning network, wherein an inclusion v2 is adopted as a feature extraction network of the network, and the target detection network is trained by utilizing data in a training set of a front target detection network. In the embodiment, in order to save time and improve efficiency, the SSD deep learning network with the highest speed is adopted, and the inclusion v2 feature extraction network is adopted to effectively improve efficiency and save time.
Training a car number recognition deep learning network by using data in a training set of the car number recognition network; the vehicle number recognition deep learning network is a faster rcnn network, wherein VGG-16 is adopted as a feature extraction network of the network. VGG-16 has higher operating efficiency relative to resnet, and has higher identification precision relative to inclusion v 2.
Step three: the collection of the image is carried out,
the method comprises the steps of collecting car number images on two sides of a vehicle to be detected, firstly, respectively carrying out projection transformation on the two car number images, and correcting image distortion caused by camera elevation shooting.
Then, the color, brightness and contrast of the two car number images are respectively adjusted, and image changes caused by different illumination conditions are corrected.
And finally, adjusting the two car number images into to-be-detected pictures with pixels of 512 multiplied by 512.
Step four: the identification of the number of the car,
the learning rate of the target detection network is set to 0.0001 by the callback parameter, and the extended anchor point is set to [0.1,0.2,0.34,0.48,0.62,0.76,0.80,0.94 ].
Because the height positions of the car numbers in the images of different car types are different, the car number images with a large range need to be acquired. But can not directly locate and classify to the smaller characters in the car number picture with a larger range. Therefore, the present embodiment adopts a policy of first locating and then identifying the car number.
And inputting the two pictures to be detected into a target detection network, wherein in the obtained result, a plurality of detection result position frames sometimes appear in the same picture, at the moment, the confidence degrees of different position frames are compared, and the position frame with the maximum confidence degree is taken as a final result, namely the position frame where the car number is located in the pictures to be detected is obtained.
And intercepting the detection picture according to the position frame where the car number is in the picture to be detected to obtain two car number screenshots, and respectively inputting the two car number screenshots into a car number recognition deep learning network to obtain the car number types corresponding to the two car number screenshots.
Step five: as a result of the determination,
because the color area-array camera is adopted to shoot the car number picture, the car number picture is easily interfered by sunlight. When sunlight is strong, the side of the vehicle body is easy to reflect light, so that picture information is not clear. Therefore, in the present embodiment, the car numbers on both sides of the vehicle body are recognized by using a mode in which two cameras simultaneously photograph the car numbers on both sides of the vehicle body. So that the car number types corresponding to the two car number screenshots can be obtained. At this time, whether the confidence degrees of the two car number types are the same or not is judged,
when the types of the car numbers on the two sides are the same, the car numbers can be used as the output result of the car numbers,
and when the car number types on the two sides are different, the car number type with high confidence coefficient is used as a car number output result. At this point, the identification of the car number is completed.
Claims (8)
1. The method for identifying the locomotive number based on deep learning is characterized by comprising the following steps of:
the method comprises the following steps: a training set is established, and the training set is established,
collecting car number pictures of different types of vehicles at two sides of the vehicle at different time and under different environments, establishing a sample library by utilizing all the car number pictures,
respectively marking the position of the car number in each car number picture, generating corresponding label files, taking all the car number pictures and the corresponding label files as a training set of the preposed target detection network,
intercepting the car number picture according to the position of the car number in each car number picture to obtain a screenshot containing the car number, marking characters of the car number in the screenshot, marking different characters as different types, and taking all the screenshots and the types corresponding to the characters in the screenshots as a training set of a car number recognition network;
step two: the training of the weight is carried out,
training the target detection network using data in the training set of the pre-target detection network,
training a car number recognition deep learning network by using data in a training set of the car number recognition network;
step three: the collection of the image is carried out,
collecting car number images on two sides of a vehicle to be detected, and adjusting the two car number images into a picture to be detected with pixels of 512 multiplied by 512;
step four: the identification of the number of the car,
inputting two pictures to be detected into a target detection network to obtain a position frame of the car number in the pictures to be detected,
intercepting the detection picture according to the position frame of the car number in the picture to be detected to obtain two car number screenshots, respectively inputting the two car number screenshots into a car number recognition deep learning network to obtain car number types corresponding to the two car number screenshots,
step five: as a result of the determination,
judging whether the confidence degrees of the two car number types are the same,
if so, the two screenshots of the car numbers have the same car number type and can be used as the output result of the car numbers,
otherwise, the type of the car number with high confidence coefficient is used as a car number output result.
2. The method for identifying a railroad locomotive number based on deep learning of claim 1, wherein the pictures of the number of the two sides of the vehicle are collected by high-definition image collecting devices disposed around the track of the vehicle.
3. The method for identifying a railroad locomotive number based on deep learning of claim 1, wherein the sample library further comprises a stretched, rotated and mirrored picture of the acquired car number picture.
4. The deep learning based recognition method of railroad locomotive number according to claim 1,
after the car number images at the two sides of the vehicle to be detected are collected, firstly, the two car number images are respectively subjected to projection transformation to correct the image distortion,
then, the color, brightness and contrast of the two car number images are respectively adjusted, the color of the images is corrected,
and finally, adjusting image pixels to obtain the picture to be detected.
5. The deep learning-based railroad locomotive number identification method according to claim 1, wherein the learning rate of the target detection network is 0.0001 and the anchor points are [0.1,0.2,0.34,0.48,0.62,0.76,0.80,0.94 ].
6. The deep learning-based railroad locomotive number identification method according to claim 1 or 4,
and when the target detection network outputs the position frames where the plurality of car numbers are located, taking the position frame with the maximum reliability as a final result.
7. The deep learning based railway locomotive number identification method according to claim 1, wherein the target detection network is an SSD deep learning network, and the Incep v2 is adopted as a feature extraction network of the network.
8. The deep learning-based railroad locomotive number identification method according to claim 1, wherein the train number identification deep learning network is a faster rcnn network, wherein VGG-16 is used as a feature extraction network of the network.
Priority Applications (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN202010260371.9A CN111539421A (en) | 2020-04-03 | 2020-04-03 | Deep learning-based railway locomotive number identification method |
Applications Claiming Priority (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN202010260371.9A CN111539421A (en) | 2020-04-03 | 2020-04-03 | Deep learning-based railway locomotive number identification method |
Publications (1)
Publication Number | Publication Date |
---|---|
CN111539421A true CN111539421A (en) | 2020-08-14 |
Family
ID=71976880
Family Applications (1)
Application Number | Title | Priority Date | Filing Date |
---|---|---|---|
CN202010260371.9A Pending CN111539421A (en) | 2020-04-03 | 2020-04-03 | Deep learning-based railway locomotive number identification method |
Country Status (1)
Country | Link |
---|---|
CN (1) | CN111539421A (en) |
Cited By (3)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN112418097A (en) * | 2020-11-24 | 2021-02-26 | 中铁第一勘察设计院集团有限公司 | Subway train number identification method based on deep learning target detection |
CN112613560A (en) * | 2020-12-24 | 2021-04-06 | 哈尔滨市科佳通用机电股份有限公司 | Method for identifying front opening and closing damage fault of railway bullet train head cover based on Faster R-CNN |
CN113920140A (en) * | 2021-11-12 | 2022-01-11 | 哈尔滨市科佳通用机电股份有限公司 | Wagon pipe cover falling fault identification method based on deep learning |
Citations (6)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN104268596A (en) * | 2014-09-25 | 2015-01-07 | 深圳市捷顺科技实业股份有限公司 | License plate recognizer and license plate detection method and system thereof |
CN104331691A (en) * | 2014-11-28 | 2015-02-04 | 深圳市捷顺科技实业股份有限公司 | Vehicle logo classifier training method, vehicle logo recognition method and device |
US20180077180A1 (en) * | 2016-09-09 | 2018-03-15 | Ca, Inc. | Bot detection system based on deep learning |
CN109325445A (en) * | 2018-09-19 | 2019-02-12 | 智慧互通科技有限公司 | A kind of method and device of Car license recognition classification |
CN110472633A (en) * | 2019-08-15 | 2019-11-19 | 南京拓控信息科技股份有限公司 | A kind of detection of train license number and recognition methods based on deep learning |
CN110796052A (en) * | 2019-10-21 | 2020-02-14 | 和昌未来科技(深圳)有限公司 | License plate recognition method, device, equipment and readable medium |
-
2020
- 2020-04-03 CN CN202010260371.9A patent/CN111539421A/en active Pending
Patent Citations (6)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN104268596A (en) * | 2014-09-25 | 2015-01-07 | 深圳市捷顺科技实业股份有限公司 | License plate recognizer and license plate detection method and system thereof |
CN104331691A (en) * | 2014-11-28 | 2015-02-04 | 深圳市捷顺科技实业股份有限公司 | Vehicle logo classifier training method, vehicle logo recognition method and device |
US20180077180A1 (en) * | 2016-09-09 | 2018-03-15 | Ca, Inc. | Bot detection system based on deep learning |
CN109325445A (en) * | 2018-09-19 | 2019-02-12 | 智慧互通科技有限公司 | A kind of method and device of Car license recognition classification |
CN110472633A (en) * | 2019-08-15 | 2019-11-19 | 南京拓控信息科技股份有限公司 | A kind of detection of train license number and recognition methods based on deep learning |
CN110796052A (en) * | 2019-10-21 | 2020-02-14 | 和昌未来科技(深圳)有限公司 | License plate recognition method, device, equipment and readable medium |
Cited By (4)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN112418097A (en) * | 2020-11-24 | 2021-02-26 | 中铁第一勘察设计院集团有限公司 | Subway train number identification method based on deep learning target detection |
CN112613560A (en) * | 2020-12-24 | 2021-04-06 | 哈尔滨市科佳通用机电股份有限公司 | Method for identifying front opening and closing damage fault of railway bullet train head cover based on Faster R-CNN |
CN113920140A (en) * | 2021-11-12 | 2022-01-11 | 哈尔滨市科佳通用机电股份有限公司 | Wagon pipe cover falling fault identification method based on deep learning |
CN113920140B (en) * | 2021-11-12 | 2022-04-19 | 哈尔滨市科佳通用机电股份有限公司 | Wagon pipe cover falling fault identification method based on deep learning |
Similar Documents
Publication | Publication Date | Title |
---|---|---|
CN111539421A (en) | Deep learning-based railway locomotive number identification method | |
CN105046196B (en) | Front truck information of vehicles structuring output method based on concatenated convolutional neutral net | |
CN112434695B (en) | Upper pull rod fault detection method based on deep learning | |
JP6904614B2 (en) | Object detection device, prediction model creation device, object detection method and program | |
CN107316001A (en) | Small and intensive method for traffic sign detection in a kind of automatic Pilot scene | |
Dehghan et al. | View independent vehicle make, model and color recognition using convolutional neural network | |
CN111191611B (en) | Traffic sign label identification method based on deep learning | |
CN111080598A (en) | Bolt and nut missing detection method for coupler yoke key safety crane | |
CN111127448B (en) | Method for detecting air spring fault based on isolated forest | |
CN109829458B (en) | Method for automatically generating log file for recording system operation behavior in real time | |
CN106355180A (en) | Method for positioning license plates on basis of combination of color and edge features | |
CN113788051A (en) | Train on-station running state monitoring and analyzing system | |
CN111382743A (en) | License plate character recognition method based on data enhancement and data generation | |
CN111091551A (en) | Method for detecting loss fault of brake beam strut opening pin of railway wagon | |
CN115063725A (en) | Airplane skin defect identification system based on multi-scale self-adaptive SSD algorithm | |
CN113657339A (en) | Instrument pointer counting and reading method and medium based on machine vision | |
CN117372677A (en) | Method for detecting health state of cotter pin of fastener of high-speed railway overhead contact system | |
CN113591643A (en) | Underground vehicle station entering and exiting detection system and method based on computer vision | |
CN117274967A (en) | Multi-mode fusion license plate recognition algorithm based on convolutional neural network | |
CN115170923B (en) | Fault identification method for loss of railway wagon supporting plate nut | |
CN110298347A (en) | A kind of recognition methods of the automobile exhaust analyzer screen based on GrayWorld and PCA-CNN | |
CN111833323B (en) | Image quality judgment method for task-divided rail wagon based on sparse representation and SVM (support vector machine) | |
CN112101107B (en) | Intelligent recognition method for intelligent network connection model vehicle on-loop simulation traffic signal lamp | |
CN111583341B (en) | Cloud deck camera shift detection method | |
CN114821498A (en) | Laser point cloud target detection and identification device and method based on deep learning |
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 | ||
RJ01 | Rejection of invention patent application after publication | ||
RJ01 | Rejection of invention patent application after publication |
Application publication date: 20200814 |