CN113052789A - Vehicle bottom plate foreign body hitting fault detection method based on deep learning - Google Patents

Vehicle bottom plate foreign body hitting fault detection method based on deep learning Download PDF

Info

Publication number
CN113052789A
CN113052789A CN202011212215.1A CN202011212215A CN113052789A CN 113052789 A CN113052789 A CN 113052789A CN 202011212215 A CN202011212215 A CN 202011212215A CN 113052789 A CN113052789 A CN 113052789A
Authority
CN
China
Prior art keywords
image
bottom plate
train
learning
sample
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
Application number
CN202011212215.1A
Other languages
Chinese (zh)
Inventor
邓艳
Current Assignee (The listed assignees may be inaccurate. Google has not performed a legal analysis and makes no representation or warranty as to the accuracy of the list.)
Harbin Kejia General Mechanical and Electrical Co Ltd
Original Assignee
Harbin Kejia General Mechanical and Electrical Co Ltd
Priority date (The priority date 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 date listed.)
Filing date
Publication date
Application filed by Harbin Kejia General Mechanical and Electrical Co Ltd filed Critical Harbin Kejia General Mechanical and Electrical Co Ltd
Priority to CN202011212215.1A priority Critical patent/CN113052789A/en
Publication of CN113052789A publication Critical patent/CN113052789A/en
Pending legal-status Critical Current

Links

Images

Classifications

    • 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
    • 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
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06F—ELECTRIC DIGITAL DATA PROCESSING
    • G06F18/00—Pattern recognition
    • G06F18/20—Analysing
    • G06F18/21—Design or setup of recognition systems or techniques; Extraction of features in feature space; Blind source separation
    • G06F18/214—Generating training patterns; Bootstrap methods, e.g. bagging or boosting
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06F—ELECTRIC DIGITAL DATA PROCESSING
    • G06F18/00—Pattern recognition
    • G06F18/20—Analysing
    • G06F18/24—Classification techniques
    • G06F18/241—Classification techniques relating to the classification model, e.g. parametric or non-parametric approaches
    • G06F18/2415—Classification techniques relating to the classification model, e.g. parametric or non-parametric approaches based on parametric or probabilistic models, e.g. based on likelihood ratio or false acceptance rate versus a false rejection rate

Landscapes

  • Engineering & Computer Science (AREA)
  • Theoretical Computer Science (AREA)
  • Physics & Mathematics (AREA)
  • Data Mining & Analysis (AREA)
  • General Physics & Mathematics (AREA)
  • Computer Vision & Pattern Recognition (AREA)
  • Artificial Intelligence (AREA)
  • Bioinformatics & Cheminformatics (AREA)
  • Bioinformatics & Computational Biology (AREA)
  • Evolutionary Biology (AREA)
  • Evolutionary Computation (AREA)
  • General Engineering & Computer Science (AREA)
  • Life Sciences & Earth Sciences (AREA)
  • Probability & Statistics with Applications (AREA)
  • Quality & Reliability (AREA)
  • Mechanical Engineering (AREA)
  • Image Analysis (AREA)

Abstract

A motor train unit bottom plate foreign body impact fault detection method based on deep learning belongs to the field of train operation fault detection. The invention aims to solve the problems of low component failure efficiency and low accuracy caused by foreign body impact on the bottom plate of a manual maintenance action vehicle set. The method comprises the following steps: establishing a train bottom plate image sample set, and acquiring a labeled image sample to be trained and an unlabeled learning image sample; carrying out sample amplification on the marked labeled image and the unmarked unlabeled image through a teacher-student model, and creating a self-classification detection model; and calculating and detecting the position of the fault through a self-classification detection model, and transmitting a message back. The method is used for fault detection of the foreign matter impact trace of the train bottom plate.

Description

Vehicle bottom plate foreign body hitting fault detection method based on deep learning
Technical Field
The invention belongs to the field of train operation fault detection, and particularly relates to a vehicle bottom plate foreign body impact fault detection method based on deep learning.
Background
The motor train unit train has the characteristics of complex structure, more parts, low bottom plate height and the like, most of the motor train unit trains are high stations in the running process, and meanwhile, the motor train unit trains have the characteristics of direct one station in the running state, return running and short stop time, and the motor train unit trains cannot manually detect and overhaul the states of the bottom plate parts midway. The damage of bottom parts caused by foreign object impact in the running process is difficult to be confirmed and detected in time, and the motor train unit running in a turn-back running mode is easy to cause faulty running. Particularly, the motor train unit running on the broken stone track is easy to cause foreign matters to hit a bottom part in the high-speed running process, and potential safety hazards of running exist. Along with the high strength use of EMUs, the mode of manual maintenance can't reach high efficiency, and high accuracy only relies on traditional manual maintenance mode and can't satisfy EMUs maintenance demand.
The deep learning and artificial intelligence technology is mature continuously, so that the foreign body hitting traces at the bottom of the motor train unit can be detected by adopting the deep learning, and the detection efficiency and the detection accuracy can be effectively improved.
Disclosure of Invention
The invention aims to solve the problems of low component failure efficiency and low accuracy rate caused by foreign body impact on the bottom plate of an action vehicle set during manual maintenance. The vehicle bottom plate foreign body hitting fault detection method based on deep learning is provided.
A vehicle bottom foreign body hitting fault detection method based on deep learning comprises the following steps:
the method comprises the following steps of firstly, obtaining a train bottom plate image sample set, wherein the train bottom plate image sample set comprises marked image samples with labels and unmarked learning image samples;
secondly, carrying out sample amplification on the marked labeled image sample and the unmarked learning image sample through a teacher-student model, establishing a self-training classification model, carrying out iterative training on the self-training classification model by using the sample obtained after amplification and taking a minimum GHM loss function as a target, and stopping iteration until the GHM loss function is completely converged;
and step three, after the iteration is stopped, the trained self-training classification model detects foreign matter striking traces of the train bottom plate image to be detected, box coordinates of the detected foreign matter striking traces are projected to coordinates on the whole train, and the box coordinates are specific positions of the foreign matter striking traces on the train bottom plate image to be detected.
Advantageous effects
1. The invention utilizes the automatic image identification mode to replace manual detection, thereby improving the detection efficiency and the accuracy.
2. According to the method, a teacher-student model mode is adopted to increase data marking samples, so that the number of the trace characteristics of foreign matters striking the train bottom plate is kept balanced relative to other characteristics in the train bottom plate image, the number of labels of a data set is increased, and the accuracy of fault identification and the generalization capability of a detection algorithm are improved.
3. Aiming at the special form of foreign matter striking marks in the train bottom plate image, a GHM loss function is adopted in the model, and the convergence rate of the model in deep learning training is increased.
4. Model training is carried out by utilizing a GHM loss function, the hit trace of the foreign matter on the bottom plate of the train is detected, difficult characteristics such as hanging of the foreign matter on the bottom plate of the train, oil leakage and water stain can be distinguished, the model is effectively prevented from being too much concerned about difficult samples in the training, and the hit trace of the foreign matter on the bottom plate is more accurately classified.
Drawings
FIG. 1 is a flow chart of the fault detection of the present invention;
FIG. 2 is a flowchart of the teacher-student model.
Detailed Description
It should be noted that, in the present invention, the embodiments disclosed in the present application may be combined with each other without conflict.
The first embodiment is as follows: the present embodiment will be described with reference to fig. 1, and the present embodiment is a vehicle underbody foreign matter impact failure detection method based on deep learning, including:
the method comprises the following steps of firstly, obtaining a train bottom plate image sample set, wherein the train bottom plate image sample set comprises marked image samples with labels and unmarked learning image samples;
secondly, carrying out sample amplification on the marked labeled image sample and the unmarked learning image sample through a teacher-student model, establishing a self-training classification model, carrying out iterative training on the self-training classification model by using the sample obtained after amplification and taking a minimum GHM loss function as a target, and stopping iteration until the GHM loss function is completely converged;
and step three, after the iteration is stopped, the trained self-training classification model detects foreign matter striking traces of the train bottom plate image to be detected, box coordinates of the detected foreign matter striking traces are projected to coordinates on the whole train, and the box coordinates are specific positions of the foreign matter striking traces on the train bottom plate image to be detected.
The second embodiment is as follows: the embodiment is different from the specific embodiment in that the first step is to obtain a car bottom plate image sample set, wherein the train bottom plate image sample set comprises a labeled image sample to be trained and an unlabeled learning image sample; the specific process is as follows:
acquiring a train bottom plate image, namely a gray level image; the bottom of the car has the characteristics of complex structure, more parts, low bottom height and the like, so that targets (foreground targets in the images, namely, bottom plate foreign matter hitting marks, water stains, oil stains, bottom plate parts and the like) in the collected bottom plate gray level images are complex, and meanwhile, the images in different weather environments can be collected in consideration of high-frequency return of the car, and the images shot by different detection stations have certain difference; compared with each part, the vent and the like, foreign body hitting traces on the bottom plate of the vehicle are small in area and few in samples; therefore, in the process of collecting the foreign body hitting marks on the vehicle bottom plate, the diversity of sample points is ensured, and the foreign body hitting marks on the vehicle bottom plate under various conditions are collected as little as possible; according to the distribution characteristics of the train bottom plate components, the train bottom plate images are divided into sections according to the number of sections of carriages, each section of the train bottom plate image is divided into a plurality of small images according to the characteristic points of main components due to the large train bottom plate, part of the small images are selected for sample marking, target characteristics which are easy to influence detection of striking traces of foreign matters on the bottom plate in the marked labeled images are uniformly marked, targets can comprise components, oil leakage, water stain and the like, a real label data set is obtained and used as a labeled image sample to be trained, and the remaining unmarked small images without labels are used as learning samples; the characteristics of the bottom plate images of each carriage are different, different bottom plate images are acquired according to different train section numbers of the train, and the characteristics of the bottom plate images are acquired.
Other steps and parameters are the same as those in the first embodiment.
The third concrete implementation mode: the embodiment is different from the first or second embodiment in that the step two pairs of marked images with labels and unmarked images without labels are subjected to sample amplification through a teacher-student model, a self-training classification model is established, iterative training is carried out on the self-training classification model by taking a minimum GHM loss function as a target, and iteration is stopped until the GHM loss function is completely converged; the specific process is as follows:
step two, marking the marked image with the label as { (x)1,y1),(x2,y2),...,(xn,yn) Record the unmarked unlabeled image as
Figure BDA0002759194740000031
For annotated tagged image dataset { (x)1,y1),(x2,y2),...,(xn,yn) Training and learning are carried out, a GHM (gradient learning mechanism) loss function is minimized, and the teacher model theta is obtained through training and learningt(semi-supervised learning model for prediction of targets in unlabeled exemplars by labeled exemplars):
Figure BDA0002759194740000032
wherein n represents the number of tagged image samples; f. ofnoisedRepresents a pair xiAdding noise and then performing a function mechanism of teacher model training; x is the number ofiRepresenting the ith labeled sample; y isiDenotes xiA corresponding label; thetatRepresents a teacher model;
step two, using a teacher model obtained by training learning to print a pseudo label on the unmarked image:
Figure BDA0002759194740000033
wherein the content of the first and second substances,
Figure BDA0002759194740000034
a pseudo label obtained by learning of an unlabeled sample is represented; m represents the total number of pseudo label samples;
Figure BDA0002759194740000035
represents an unlabeled sample;
step two, using the marked data set with the label and the unmarked data set, enabling a GHM (gradient matching mechanism) loss function to be minimum through learning training, training the noise-added student model (a training model for predicting a real label of the unmarked sample through the marked sample and the label corresponding to the sample by guiding the unmarked sample through a teacher model), and iteratively training the teacher model through the following formula until the loss value l of the GHM loss functionGHMAt a minimum, the iteration is stopped:
Figure BDA0002759194740000041
other steps and parameters are the same as those in the first or second embodiment.
The fourth concrete implementation mode: the difference between this embodiment and the first to the third embodiment is that the loss function l in the second stepGHMComprises the following steps:
Figure BDA0002759194740000042
wherein p isiLabel y for representing imageiProbability of foreign object impact trace failure; p is a radical ofi *Representing a genuine label, pi *＝{0,1}；giRepresents the gradient mode length of the ith sample; GD (g)i) The gradient density is expressed and is the number of samples of a unit gradient module length g part; l isceRepresenting a cross entropy loss function; use ofGHMThe loss function can prevent the model from paying more attention to particularly difficultly distinguished samples in the vehicle bottom image, prevent misjudgment of difficultly distinguished sample discrete points when the model is converged, and pay attention to the modelThe discrete points increase the training cost, so that the training efficiency is improved, the model can be well converged, and the accuracy of detection and identification is improved.
Other steps and parameters are the same as those in one of the first to third embodiments.
The fifth concrete implementation mode: the difference between this embodiment and one of the first to the fourth embodiments is that the cross entropy loss function LceComprises the following steps:
Figure BDA0002759194740000043
other steps and parameters are the same as in one of the first to fourth embodiments.
The sixth specific implementation mode: this embodiment is different from one of the first to fifth embodiments in that the gradient density GD (g)i) Comprises the following steps:
Figure BDA0002759194740000044
wherein lεIs a cross-over loss function; deltaε(gk,gi) Denotes the gradient mode length in samples 1 to n
Figure BDA0002759194740000045
Number of samples in the range,/ε(gi) Represents a section
Figure BDA0002759194740000046
Wherein ε represents the width of each bin (interval) when the gradient is calculated; gkRepresents the gradient mode length of the kth sample; giIs the unit gradient mode length.
Other steps and parameters are the same as those in one of the first to fifth embodiments.
The seventh embodiment: the present embodiment is different from the first to sixth embodiments in that the unit gradient mode length giExpressed as:
Figure BDA0002759194740000051
wherein p represents the probability of the foreign matter impact trace failure predicted by the teacher model; p is a radical of*Representing a genuine label, p*＝{0,1}；giProportional to the difficulty of detection, the greater the difficulty of detection, giThe larger.
Other steps and parameters are the same as those in one of the first to sixth embodiments.
The specific implementation mode is eight: the second step, the third step, different from the first to the seventh embodiments, trains the noise-added student model by using the labeled dataset and the unlabeled dataset through learning training to minimize the ghm (learning modeling) loss function; the noise includes: data Augmentation (Data enhancement, enhancement by means of brightness, inversion and the like), Dropout (randomly discarding some samples according to a set threshold proportion in the training process), and Stochastic Depth (randomly discarding some training layer numbers of the teacher-student in the training process); data Augmentation input noise is added to enhance the images at the bottom of the car, so that the converted images still have correct classification labels, and the purpose of enabling the student model to predict the characteristic images of difficult samples more easily is achieved; while training the student model, adding Dropout (random discard) and Stochasic Depth (random Depth) noises to the marked labeled image samples and the unmarked learning image samples marked with pseudo labels, so that the student model is not single and the complexity is gradually close to that of the teacher model set; each class of labeled and unlabeled image samples is gradually brought into equilibrium.
Other steps and parameters are the same as those in one of the first to seventh embodiments.

Claims (9)

1. A vehicle bottom foreign body hitting fault detection method based on deep learning is characterized by comprising the following steps:
the method comprises the following steps of firstly, obtaining a train bottom plate image sample set, wherein the train bottom plate image sample set comprises marked image samples with labels and unmarked learning image samples;
secondly, carrying out sample amplification on the marked labeled image sample and the unmarked learning image sample through a teacher-student model, establishing a self-training classification model, carrying out iterative training on the self-training classification model by using the sample obtained after amplification and taking a minimum GHM loss function as a target, and stopping iteration until the GHM loss function is completely converged;
and step three, after the iteration is stopped, the trained self-training classification model detects foreign matter striking traces of the train bottom plate image to be detected, box coordinates of the detected foreign matter striking traces are projected to coordinates on the whole train, and the box coordinates are specific positions of the foreign matter striking traces on the train bottom plate image to be detected.
2. The method for detecting the foreign body hitting fault of the underbody on the basis of the deep learning as claimed in claim 1, wherein in the first step, a train underbody image sample set is established, namely, a labeled image sample to be trained and an unlabeled learning image sample are obtained; the specific process is as follows:
acquiring a train bottom plate image; dividing the images of the bottom plate of the train into sections according to the number of the sections of the train, dividing the images of the bottom plate of each section of the train into a plurality of small images, selecting partial small images for marking, wherein the marked content comprises the following steps: foreign objects hit the traces to obtain marked labeled image samples, leaving unmarked unlabeled small images as learning image samples.
3. The vehicle underbody foreign matter impact fault detection method based on deep learning of claim 2, wherein the content of the mark further comprises at least one of the following items: oil leakage, water stain, easily hit the part that the trace detected and cause the influence to the foreign matter.
4. The vehicle bottom plate foreign matter impact fault detection method based on deep learning of claim 2 is characterized in that in the step two pairs of marked labeled image samples and unmarked learning image samples are subjected to sample amplification through a teacher-student model, a self-training classification model is established, iterative training is carried out on the self-training classification model by using the samples obtained after amplification with a minimum GHM loss function as a target, and the iteration is stopped until the GHM loss function is completely converged; the specific process is as follows:
step two, marking the marked image with the label as { (x)1,y1),(x2,y2),...,(xn,yn) Record the unmarked unlabeled image as
Figure FDA0002759194730000011
Training an initial teacher model through a marked tagged data set image to obtain the teacher model: thetatWherein the loss value of the loss function for training is calculated by the following formula:
Figure FDA0002759194730000012
wherein n represents the number of tagged image samples; f. ofnoisedRepresents a pair xiA function mechanism for teacher mode training is carried out after noise is added; x is the number ofiRepresenting the ith labeled sample; y isiDenotes xiA corresponding label; thetatRepresents a teacher model;
step two, using a teacher model obtained by training learning to mark a pseudo label on an unmarked learning image sample:
Figure FDA0002759194730000021
wherein the content of the first and second substances,
Figure FDA0002759194730000022
a pseudo label obtained by learning and representing an unmarked learning image sample; m represents a pseudo labelTotal number of samples;
Figure FDA0002759194730000023
representing unlabeled learning image samples;
step two, adding noise to the marked image sample with the label and the unmarked learning image sample marked with the pseudo label;
training the teacher model iteratively until the loss value l of the GHM loss functionGHMAt a minimum, the iteration is stopped:
Figure FDA0002759194730000024
5. the vehicle underbody foreign matter hitting fault detection method based on deep learning of claim 4, wherein the GHM loss function l isGHMComprises the following steps:
Figure FDA0002759194730000025
wherein p isiLabel y for representing imageiProbability of foreign object impact trace failure; GD (g)i) The gradient density is expressed and is the number of samples of a unit gradient module length g part; l isceRepresenting a cross entropy loss function.
6. The vehicle bottom plate foreign body hitting fault detection method based on deep learning of claim 5, wherein the cross entropy loss function L isceComprises the following steps:
Figure FDA0002759194730000026
7. the vehicle bottom foreign body hitting fault detection method based on deep learning of claim 5, wherein the gradient density is calculated by the following formula:
Figure FDA0002759194730000027
wherein, deltaε(gkG) shows that the gradient mode length is in samples 1 to n
Figure FDA0002759194730000028
Number of samples in the range,/ε(g) Represents a section
Figure FDA0002759194730000031
G is the unit gradient mode length; ε represents the width of each bin when the gradient is calculated; gkRepresenting the gradient mode length of the kth sample.
8. The vehicle bottom foreign body hitting fault detection method based on deep learning as claimed in claim 5 or 7, wherein the unit gradient module length g is as follows:
Figure FDA0002759194730000032
wherein p represents the probability of the foreign matter impact trace failure predicted by the teacher model; p is a radical of*Representing a genuine label, p*＝{0,1}。
9. The deep learning-based underbody foreign matter impact fault detection method as claimed in claim 4, wherein said two or three steps add noise to the train floor image, said noise including Data Augmentation, Dropout and Stochastic Depth.
CN202011212215.1A 2020-11-03 2020-11-03 Vehicle bottom plate foreign body hitting fault detection method based on deep learning Pending CN113052789A (en)

Priority Applications (1)

Application Number Priority Date Filing Date Title
CN202011212215.1A CN113052789A (en) 2020-11-03 2020-11-03 Vehicle bottom plate foreign body hitting fault detection method based on deep learning

Applications Claiming Priority (1)

Application Number Priority Date Filing Date Title
CN202011212215.1A CN113052789A (en) 2020-11-03 2020-11-03 Vehicle bottom plate foreign body hitting fault detection method based on deep learning

Publications (1)

Publication Number Publication Date
CN113052789A true CN113052789A (en) 2021-06-29

Family

ID=76507801

Family Applications (1)

Application Number Title Priority Date Filing Date
CN202011212215.1A Pending CN113052789A (en) 2020-11-03 2020-11-03 Vehicle bottom plate foreign body hitting fault detection method based on deep learning

Country Status (1)

Country Link
CN (1) CN113052789A (en)

Cited By (1)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
WO2023077821A1 (en) * 2021-11-07 2023-05-11 西北工业大学 Multi-resolution ensemble self-training-based target detection method for small-sample low-quality image

Citations (3)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN109977817A (en) * 2019-03-14 2019-07-05 南京邮电大学 EMU car bed bolt fault detection method based on deep learning
CN111639215A (en) * 2020-06-01 2020-09-08 南京索安电子有限公司 Vehicle bottom scanning system for image learning and image recognition of artificial intelligence and implementation method
CN111814867A (en) * 2020-07-03 2020-10-23 浙江大华技术股份有限公司 Defect detection model training method, defect detection method and related device

Patent Citations (3)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN109977817A (en) * 2019-03-14 2019-07-05 南京邮电大学 EMU car bed bolt fault detection method based on deep learning
CN111639215A (en) * 2020-06-01 2020-09-08 南京索安电子有限公司 Vehicle bottom scanning system for image learning and image recognition of artificial intelligence and implementation method
CN111814867A (en) * 2020-07-03 2020-10-23 浙江大华技术股份有限公司 Defect detection model training method, defect detection method and related device

Non-Patent Citations (2)

* Cited by examiner, † Cited by third party
Title
BUYU LI等: "Gradient Harmonized Single-stage Detector", 《ARXIV》 *
QIZHE XIE等: "Self-training with Noisy Student improves ImageNet classification", 《ARXIV》 *

Cited By (1)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
WO2023077821A1 (en) * 2021-11-07 2023-05-11 西北工业大学 Multi-resolution ensemble self-training-based target detection method for small-sample low-quality image

Similar Documents

Publication Publication Date Title
CN111080598B (en) Bolt and nut missing detection method for coupler yoke key safety crane
He et al. Obstacle detection of rail transit based on deep learning
CN115272334A (en) Method for detecting micro defects on surface of steel rail under complex background
CN113947590B (en) Surface defect detection method based on multi-scale attention guidance and knowledge distillation
CN111652227B (en) Method for detecting damage fault of bottom floor of railway wagon
CN111080608A (en) Method for recognizing closing fault image of automatic brake valve plug handle of railway wagon in derailment
CN103559237B (en) Semi-automatic image annotation sample generating method based on target tracking
CN110555475A (en) few-sample target detection method based on semantic information fusion
CN111091558B (en) Railway wagon swing bolster spring jumping fault image identification method
CN111080602B (en) Method for detecting foreign matters in water leakage hole of railway wagon
CN111476756A (en) Method for identifying casting DR image loose defects based on improved YO L Ov3 network model
CN111080611A (en) Railway wagon bolster spring fracture fault image identification method
CN105809198A (en) SAR image target recognition method based on deep belief network
CN104299243B (en) Target tracking method based on Hough forests
CN106683119A (en) Moving vehicle detecting method based on aerially photographed video images
CN111080612B (en) Truck bearing damage detection method
CN111862029A (en) Fault detection method for bolt part of vertical shock absorber of railway motor train unit
CN103413149B (en) Method for detecting and identifying static target in complicated background
CN113052789A (en) Vehicle bottom plate foreign body hitting fault detection method based on deep learning
CN110717464A (en) Intelligent railway roadbed disease identification method based on radar data
CN113506268B (en) Semi-supervised foreign matter detection method for rail image
CN114549933A (en) Countermeasure sample generation method based on target detection model feature vector migration
CN106651906A (en) Test method for motion characteristic of high-voltage breaker based on improved Gaussian mixture model
CN108648210B (en) Rapid multi-target detection method and device under static complex scene
CN111275684A (en) Strip steel surface defect detection method based on multi-scale feature extraction

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

Application publication date: 20210629

RJ01 Rejection of invention patent application after publication