CN112949688A - Motor train unit bottom plate rubber damage fault detection method, system and device - Google Patents
Motor train unit bottom plate rubber damage fault detection method, system and device Download PDFInfo
- Publication number
- CN112949688A CN112949688A CN202110138125.0A CN202110138125A CN112949688A CN 112949688 A CN112949688 A CN 112949688A CN 202110138125 A CN202110138125 A CN 202110138125A CN 112949688 A CN112949688 A CN 112949688A
- Authority
- CN
- China
- Prior art keywords
- model
- plate rubber
- bottom plate
- train unit
- motor train
- 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
Images
Classifications
-
- 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
- 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
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/02—Neural networks
- G06N3/08—Learning methods
-
- 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
-
- 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]
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
- G06V2201/00—Indexing scheme relating to image or video recognition or understanding
- G06V2201/07—Target detection
Abstract
A method, a system and a device for detecting a motor train unit bottom plate rubber damage fault belong to the technical field of image detection. The method aims to solve the problems of long training time and low efficiency in the existing method for detecting the damage fault of the rubber sheet by utilizing a deep learning method. The method comprises the steps of acquiring an image of a region of interest of the base plate rubber to be detected as model input, and detecting the damage fault of the base plate rubber by using an expandable target detection model; training the extensible target detection model by using a training set in the training process of the extensible target detection model to obtain model parameter weight; performing model test by using the test set to obtain a model test recognition rate, adjusting model hyper-parameters, and repeating training and testing until the model test recognition rate reaches a test recognition rate threshold value to obtain an optimal parameter weight of the model; and the model loaded with the optimal parameter weight is the trained extensible target detection model. The method is suitable for fault detection of the damage of the rubber of the bottom plate.
Description
Technical Field
The invention relates to a method, a system and a device for detecting a motor train unit bottom plate rubber damage fault. Belonging to the technical field of image detection.
Background
The bottom plate rubber can guarantee the waterproof nature of EMUs bottom plate in high-speed operation, and the protection EMUs bottom plate inner part does not receive the influence of sewage and takes place faults such as short circuit. Therefore, the automatic alarm device has important significance in timely and automatically alarming the damage fault of the rubber of the motor train unit bottom plate.
The motor train unit bottom plate rubber damage fault is automatically detected and alarmed through a deep learning method, and only a small number of alarm results need to be confirmed manually. For all bottom plate rubber images of manual inspection, can effectively improve detection quality and detection efficiency, practice thrift the human cost of vehicle section by a wide margin. However, because the damaged failure mode of the rubber of the base plate is variable, a new failure mode is often required to be added into the training image data set, and the whole training image data set is used for retraining the target detection model, so that the model training consumes more time, and the efficiency is reduced.
Disclosure of Invention
The invention aims to solve the problems of long training time and low efficiency when the existing deep learning method is used for detecting the damage fault of the veneer.
A method for detecting a motor train unit bottom plate rubber breakage fault comprises the following steps:
acquiring an image of a region of interest of the base plate rubber to be detected as model input, and detecting the damage fault of the base plate rubber by using an expandable target detection model; the expandable target detection model comprises a plurality of backbone networks for extracting features, and each backbone network is followed by a detector for detecting a target label, confidence and position;
the extensible target detection model is trained in advance, and the specific training process comprises the following steps:
acquiring an interesting region image containing a bottom plate rubber for training, constructing a data set, and marking the data set to obtain a label file corresponding to each image; labels used for labeling are divided into: normal rubber and fault form i;
dividing all images and all corresponding labels into a training set and a testing set; training an extensible target detection model by using a training set to obtain model parameter weights; performing model test by using the test set to obtain a model test recognition rate, adjusting model hyper-parameters, and repeating training and testing until the model test recognition rate reaches a test recognition rate threshold value to obtain an optimal parameter weight of the model; and the model loaded with the optimal parameter weight is the trained extensible target detection model.
Further, the establishing process of the extensible target detection model comprises the following steps:
simplifying the bottom plate rubber region-of-interest images of known types into known type images, and respectively inputting the n types of known type images into n trunk networks for feature extraction to obtain n types of data features corresponding to the n types of known type images; inputting the n types of data characteristics into n corresponding detectors respectively for target detection to obtain labels, confidence degrees and positions of targets in n types of known images; respectively packaging the backbone network and the corresponding detectors into sub-models, and forming an integrated model by all the sub-models; simplifying a bottom plate rubber interesting area image for testing as a test image, taking a single test image as input, and simultaneously carrying out target detection on the test image through a sub-model in an integrated model to obtain target labels, confidence degrees and positions corresponding to a plurality of categories; according to the confidence threshold, screening the confidence corresponding to each of the multiple categories, and removing m categories with the confidence lower than the threshold; and taking all target labels and corresponding positions contained in the rest detection categories as a final detection result of the model, wherein the obtained model is the extensible target detection model.
Further, in the process of respectively packaging the backbone network and the corresponding detectors into sub-models and forming all the sub-models into an integrated model, when the requirement of newly added detection types exists, respectively inputting n + l known type images into n + l backbone networks for feature extraction for l newly added type images to obtain n + l type data features corresponding to the n + l known type images; respectively inputting the n + l data characteristics into corresponding n + l detectors for target detection to obtain the label, confidence coefficient and position of the target in the n + l known type images; and respectively packaging the backbone network and the corresponding detectors into sub-models, and forming an integrated model by all the sub-models.
Further, the test recognition threshold is 100%.
Further, the process of acquiring the region of interest of the base plate rubber to be detected comprises the following steps:
firstly, acquiring an integral image of the bottom of the motor train unit;
and then, according to the wheel base information between the motor train unit axles and the prior information of the position of the bottom plate rubber relative to the axles, roughly positioning the bottom plate rubber on the whole bottom image of the motor train unit to obtain an interested area image containing the bottom plate rubber.
Further, in the process of dividing all the images and all the corresponding labels into the training set and the test set, the ratio of the number of the images and the corresponding labels in the training set to the number of the corresponding labels in the test set is 80%: 20 percent.
Further, the backbone network is VGG or ResNet.
Further, the detector is fast-RCNN or YOLO.
A motor train unit bottom plate rubber breakage fault detection system is used for executing a motor train unit bottom plate rubber breakage fault detection method.
A motor train unit bottom plate rubber breakage fault detection device is used for storing and/or operating a motor train unit bottom plate rubber breakage fault detection system.
Has the advantages that:
(1) the invention provides an extensible target detection model EODM, which can be used for carrying out rapid model training aiming at newly added fault forms, simultaneously keeping the detection capability of the target detection model on the existing fault forms and improving the model training efficiency;
(2) according to the method, the deep learning method is adopted to replace manual work to carry out automatic fault detection on the motor train unit, the influence of subjective factors of detection personnel and the limitation of working time are avoided, and the detection quality and the detection efficiency of the motor train unit base plate rubber damage fault can be effectively improved.
Drawings
FIG. 1 is a schematic flow diagram of an embodiment
FIG. 2 is a schematic flow chart of the embodiment;
fig. 3 is a schematic structural diagram of an extensible target detection 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 is specifically described with reference to figure 1,
the embodiment is a motor train unit bottom plate rubber damage fault detection method, which comprises the following steps:
acquiring an image of a region of interest of the base plate rubber to be detected as model input, and detecting the damage fault of the base plate rubber by using an expandable target detection model; the expandable target detection model comprises a plurality of backbone networks for extracting features, and each backbone network is followed by a detector for detecting a target label, confidence and position;
the extensible target detection model is trained in advance, and the specific training process comprises the following steps:
acquiring an interesting region image containing a bottom plate rubber for training, constructing a data set, and marking the data set to obtain a label file corresponding to each image; labels used for labeling are divided into: normal rubber and fault form i, i is 1,2, … …, n, n is fault category number;
dividing all images and all corresponding labels into a training set and a testing set; training an extensible target detection model by using a training set to obtain model parameter weights; performing model test by using the test set to obtain a model test recognition rate, adjusting model hyper-parameters, and repeating training and testing until the model test recognition rate reaches a test recognition rate threshold value to obtain an optimal parameter weight of the model; and the model loaded with the optimal parameter weight is the trained extensible target detection model.
The second embodiment is as follows:
the embodiment is a method for detecting a motor train unit bottom plate rubber damage fault, and the establishment process of the extensible target detection model in the embodiment comprises the following steps:
simplifying the bottom plate rubber region-of-interest images of known types into known type images, and respectively inputting the n types of known type images into n trunk networks for feature extraction to obtain n types of data features corresponding to the n types of known type images; inputting the n types of data characteristics into n corresponding detectors respectively for target detection to obtain labels, confidence degrees and positions of targets in n types of known images; respectively packaging the backbone network and the corresponding detectors into sub-models, and forming an integrated model by all the sub-models; simplifying a bottom plate rubber interesting area image for testing as a test image, taking a single test image as input, and simultaneously carrying out target detection on the test image through a sub-model in an integrated model to obtain target labels, confidence degrees and positions corresponding to a plurality of categories; according to the confidence threshold, screening the confidence corresponding to each of the multiple categories, and removing m categories with the confidence lower than the threshold; and taking all target labels and corresponding positions contained in the rest detection categories as a final detection result of the model, wherein the obtained model is the extensible target detection model.
Other steps and parameters are the same as in the first embodiment.
The third concrete implementation mode:
in the embodiment, in the process of respectively packaging a trunk network and corresponding detectors into submodels and forming all the submodels into an integrated model, when the requirement of newly added detection types exists, for l newly added type images, respectively inputting n + l known type images into n + l trunk networks for feature extraction to obtain n + l type data features corresponding to the n + l known type images; respectively inputting the n + l data characteristics into corresponding n + l detectors for target detection to obtain the label, confidence coefficient and position of the target in the n + l known type images; and respectively packaging the backbone network and the corresponding detectors into sub-models, and forming an integrated model by all the sub-models.
Other steps and parameters are the same as in the second embodiment.
The fourth concrete implementation mode:
the embodiment is a method for detecting a damage fault of a rubber sheet of a motor train unit bottom plate, and the threshold value of the test recognition rate is 100%.
Other steps and parameters are the same as in one of the first to third embodiments.
The fifth concrete implementation mode:
the embodiment is a method for detecting a damage fault of a motor train unit base plate rubber sheet, and the process for acquiring a base plate rubber sheet interesting area to be detected in the embodiment comprises the following steps:
firstly, acquiring an integral image of the bottom of the motor train unit;
and then, according to the wheel base information between the motor train unit axles and the prior information of the position of the bottom plate rubber relative to the axles, roughly positioning the bottom plate rubber on the whole bottom image of the motor train unit to obtain an interested area image containing the bottom plate rubber.
Other steps and parameters are the same as in one of the first to fourth embodiments.
The sixth specific implementation mode:
in the embodiment, in the process of dividing all images and corresponding labels into a training set and a testing set, the ratio of the number of the images and the corresponding labels in the training set and the testing set is 80%: 20 percent.
Other steps and parameters are the same as in one of the first to fifth embodiments.
The seventh embodiment:
the embodiment is a motor train unit bottom plate rubber breakage fault detection method, and a main network in the embodiment is VGG or ResNet.
Other steps and parameters are the same as in one of the first to sixth embodiments.
The specific implementation mode is eight:
the embodiment is a motor train unit bottom plate rubber breakage fault detection method, and a detector in the embodiment is fast-RCNN or YOLO.
Other steps and parameters are the same as in one of the first to seventh embodiments.
The specific implementation method nine:
the embodiment is a motor train unit bottom plate rubber breakage fault detection system, which is used for executing a motor train unit bottom plate rubber breakage fault detection method.
The detailed implementation mode is ten:
the embodiment is a motor train unit bottom plate rubber breakage fault detection device which is used for storing and/or operating a motor train unit bottom plate rubber breakage fault detection system.
Examples
The embodiment is a method for detecting a motor train unit bottom plate rubber breakage fault, as shown in fig. 2, and the method comprises the following steps:
step 1.1, carrying a camera by using a fixed device installed beside a track, calculating the shooting frequency of the camera according to the moving speed of the motor train unit, and continuously shooting the motor train unit;
and step 1.2, splicing the plurality of images shot in the step 1.1 to obtain an integral image of the bottom of the motor train unit.
and acquiring wheelbase information between the axles of the motor train unit by using a sensor arranged beside the track. According to the wheel base information and the prior information of the approximate position of the bottom plate rubber relative to the axle, the bottom plate rubber is roughly positioned on the whole bottom image of the motor train unit obtained in the step 1.2, and an interested area image containing the bottom plate rubber is obtained.
Step 3, preprocessing the data set:
3.1, manually screening an image containing normal bottom plate rubber and an image containing damaged bottom plate rubber from the image of the region of interest obtained in the step 2, and dividing the images into two types so as to construct an original image data set;
step 3.2, because the number of the damaged bottom plate rubber images obtained in the step 3.1 is usually less, data amplification operations such as brightness conversion, histogram equalization and the like are carried out on the damaged bottom plate rubber images so as to improve the training effect of the model;
and 3.3, labeling all images by using a labelImg data labeling tool, wherein labels used for labeling are as follows: obtaining a label file corresponding to each image by analogy with the normal rubber, the fault form 1 and the fault form 2;
and 3.4, segmenting all the images and all the corresponding labels, wherein 80% of the images and the corresponding labels are divided into a training data set. The remaining 20% of the image and corresponding label are divided into test datasets.
And 4, constructing an extensible target detection model, wherein the structure of the extensible target detection model provided by the invention is shown in FIG. 3. The scalable object detection model includes a plurality of backbone networks and equal number of detectors, and the following process will be described in order to describe the structure of the scalable object detection model, which is only for describing the data input method and does not actually start inputting data and training.
And 4.1, respectively inputting the n kinds of known class images into the n trunk networks for feature extraction to obtain n kinds of data features corresponding to the n kinds of known class images. Wherein, the backbone network can select backbone networks widely used in the field such as VGG, ResNet, etc.;
step 4.2, inputting the n-type data characteristics obtained in the step 4.1 into corresponding n detectors respectively for target detection to obtain labels, confidence degrees and positions of targets in n types of known images; wherein, the detector can be a target detector widely used in the field, such as fast-RCNN, YOLO, etc.;
step 4.3, when the requirement of the newly added detection type exists, processing the 1 newly added type images in the mode of the step 4.1 to the step 4.2 to obtain the label, the confidence coefficient and the position of the target in the 1 newly added type images, and realizing the expandability of the model;
and 4.4, respectively packaging the n +1 trunk networks and the corresponding detectors in the steps 4.1-4.3 into n +1 submodels, and forming an integrated model by all the submodels. The integrated model can take a single test image as input, and simultaneously carry out target detection on the test image through n +1 sub-models inside to obtain target labels, confidence degrees and positions corresponding to n +1 categories;
step 4.5, according to the manually set confidence threshold, screening the confidence corresponding to each of the n +1 categories obtained in the step 4.4, and removing m categories with the confidence lower than the threshold; and taking all target labels and corresponding positions contained in the remaining n +1-m detection categories as a final detection result of the model, wherein the obtained model is the extensible target detection model. Since there may be a plurality of classes of objects in the same image, this step simultaneously outputs the detection objects included in the remaining n-m detection classes as the final detection result.
In fact, the main differences between the present invention and the prior art are as follows: the invention constructs the submodel according to the target label category, but not according to different network structures. The method has the advantages that when a new fault type needs to be detected, only data corresponding to the new fault type needs to be used for training a new sub-model and is added into the existing integrated model, and the whole integrated model does not need to be retrained by using all types of data.
Research and experiments show that in the detection of the damage of the rubber of the bottom plate, training samples of the existing fault types are basically not required to be added, so that the processing mode is selected, and meanwhile, a sub-model is independently constructed for each fault type, and the detection performance of the model can be improved.
Step 5, model training and testing:
step 5.1, training the extensible target detection model by utilizing the training data set obtained in the step 3.4 according to the manually set model hyper-parameters to obtain model parameter weight;
step 5.2, loading the obtained model parameter weight into an expandable target detection model, and performing model test on the test data set obtained in the step 3.4 to obtain a model test recognition rate;
step 5.3, adjusting the model hyper-parameters, and repeating the step 5.1 to the step 5.2 until the model test recognition rate reaches 100 percent to obtain the optimal parameter weight of the model;
step 6, using the extensible target detection model to carry out actual detection:
loading the optimal parameter weight of the model obtained in the step 5.3 into an expandable target detection model; using the images of the region of interest of the base plate rubber to be detected obtained in the steps 1-2 as model input for detection; if the bottom plate rubber damage fault is detected in the image of the region of interest, the fault position is automatically uploaded to an alarm platform.
The present invention is capable of other embodiments and its several details are capable of modifications in various obvious respects, all without departing from the spirit and scope of the present invention.
Claims (10)
1. A motor train unit bottom plate rubber breakage fault detection method is characterized by comprising the following steps:
acquiring an image of a region of interest of the base plate rubber to be detected as model input, and detecting the damage fault of the base plate rubber by using an expandable target detection model; the expandable target detection model comprises a plurality of backbone networks for extracting features, and each backbone network is followed by a detector for detecting a target label, confidence and position;
the extensible target detection model is trained in advance, and the specific training process comprises the following steps:
acquiring an interesting region image containing a bottom plate rubber for training, constructing a data set, and marking the data set to obtain a label file corresponding to each image; labels used for labeling are divided into: normal rubber and fault form i;
dividing all images and all corresponding labels into a training set and a testing set; training an extensible target detection model by using a training set to obtain model parameter weights; performing model test by using the test set to obtain a model test recognition rate, adjusting model hyper-parameters, and repeating training and testing until the model test recognition rate reaches a test recognition rate threshold value to obtain an optimal parameter weight of the model; and the model loaded with the optimal parameter weight is the trained extensible target detection model.
2. The method for detecting the motor train unit bottom plate rubber breakage fault as claimed in claim 1, wherein the process of establishing the expandable target detection model comprises the following steps:
simplifying the bottom plate rubber region-of-interest images of known types into known type images, and respectively inputting the n types of known type images into n trunk networks for feature extraction to obtain n types of data features corresponding to the n types of known type images; inputting the n types of data characteristics into n corresponding detectors respectively for target detection to obtain labels, confidence degrees and positions of targets in n types of known images; respectively packaging the backbone network and the corresponding detectors into sub-models, and forming an integrated model by all the sub-models; simplifying a bottom plate rubber interesting area image for testing as a test image, taking a single test image as input, and simultaneously carrying out target detection on the test image through a sub-model in an integrated model to obtain target labels, confidence degrees and positions corresponding to a plurality of categories; according to the confidence threshold, screening the confidence corresponding to each of the multiple categories, and removing m categories with the confidence lower than the threshold; and taking all target labels and corresponding positions contained in the rest detection categories as a final detection result of the model, wherein the obtained model is the extensible target detection model.
3. The method for detecting the motor train unit bottom plate rubber breakage fault of claim 2, wherein in the process of respectively packaging the main network and the corresponding detectors into the sub-models and forming all the sub-models into an integrated model, when a requirement of a newly added detection type exists, for images of l newly added types, n + l known type images are respectively input into n + l main networks for feature extraction, and n + l type data features corresponding to the n + l known type images are obtained; respectively inputting the n + l data characteristics into corresponding n + l detectors for target detection to obtain the label, confidence coefficient and position of the target in the n + l known type images; and respectively packaging the backbone network and the corresponding detectors into sub-models, and forming an integrated model by all the sub-models.
4. The method for detecting the breakage fault of the rubber of the motor train unit bottom plate as claimed in claim 3, wherein the test recognition rate threshold is 100%.
5. The motor train unit base plate rubber breakage fault detection method as claimed in claim 1,2, 3 or 4, characterized in that the process of obtaining the base plate rubber region of interest to be detected comprises the following steps:
firstly, acquiring an integral image of the bottom of the motor train unit;
and then, according to the wheel base information between the motor train unit axles and the prior information of the position of the bottom plate rubber relative to the axles, roughly positioning the bottom plate rubber on the whole bottom image of the motor train unit to obtain an interested area image containing the bottom plate rubber.
6. The method for detecting the motor train unit bottom plate rubber breakage fault of claim 5, wherein in the process of dividing all the images and corresponding labels into a training set and a testing set, the ratio of the number of the images and corresponding labels in the training set to the number of the labels in the testing set is 80%: 20 percent.
7. The method for detecting the motor train unit base plate rubber breakage fault of claim 6, wherein the main network is VGG or ResNet.
8. The method as claimed in claim 7, wherein the detector is fast-RCNN or YOLO.
9. A motor train unit base plate rubber breakage fault detection system, characterized in that the system is used for executing the motor train unit base plate rubber breakage fault detection method as claimed in one of claims 1 to 8.
10. The motor train unit bottom plate rubber breakage fault detection device is characterized by being used for storing and/or operating the motor train unit bottom plate rubber breakage fault detection system as claimed in claim 9.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202110138125.0A CN112949688A (en) | 2021-02-01 | 2021-02-01 | Motor train unit bottom plate rubber damage fault detection method, system and device |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202110138125.0A CN112949688A (en) | 2021-02-01 | 2021-02-01 | Motor train unit bottom plate rubber damage fault detection method, system and device |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| CN112949688A true CN112949688A (en) | 2021-06-11 |
Family
ID=76241131
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202110138125.0A Pending CN112949688A (en) | 2021-02-01 | 2021-02-01 | Motor train unit bottom plate rubber damage fault detection method, system and device |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN112949688A (en) |
Citations (5)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN109344905A (en) * | 2018-10-22 | 2019-02-15 | 王子蕴 | A kind of transmission facility automatic fault recognition methods based on integrated study |
| CN110163234A (en) * | 2018-10-10 | 2019-08-23 | 腾讯科技（深圳）有限公司 | A kind of model training method, device and storage medium |
| US20200082165A1 (en) * | 2016-12-16 | 2020-03-12 | Peking University Shenzhen Graduate School | Collaborative deep network model method for pedestrian detection |
| CN111862067A (en) * | 2020-07-28 | 2020-10-30 | 中山佳维电子有限公司 | Welding defect detection method and device, electronic equipment and storage medium |
| CN112257808A (en) * | 2020-11-02 | 2021-01-22 | 郑州大学 | Integrated collaborative training method and device for zero sample classification and terminal equipment |
-
2021
- 2021-02-01 CN CN202110138125.0A patent/CN112949688A/en active Pending
Patent Citations (5)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US20200082165A1 (en) * | 2016-12-16 | 2020-03-12 | Peking University Shenzhen Graduate School | Collaborative deep network model method for pedestrian detection |
| CN110163234A (en) * | 2018-10-10 | 2019-08-23 | 腾讯科技（深圳）有限公司 | A kind of model training method, device and storage medium |
| CN109344905A (en) * | 2018-10-22 | 2019-02-15 | 王子蕴 | A kind of transmission facility automatic fault recognition methods based on integrated study |
| CN111862067A (en) * | 2020-07-28 | 2020-10-30 | 中山佳维电子有限公司 | Welding defect detection method and device, electronic equipment and storage medium |
| CN112257808A (en) * | 2020-11-02 | 2021-01-22 | 郑州大学 | Integrated collaborative training method and device for zero sample classification and terminal equipment |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN111507990B (en) | Tunnel surface defect segmentation method based on deep learning | |
| CN111091555B (en) | Brake shoe breaking target detection method | |
| CN110569843B (en) | Intelligent detection and identification method for mine target | |
| EP3690741A2 (en) | Method for automatically evaluating labeling reliability of training images for use in deep learning network to analyze images, and reliability-evaluating device using the same | |
| CN111754498A (en) | Conveyor belt carrier roller detection method based on YOLOv3 | |
| CN110210433B (en) | Container number detection and identification method based on deep learning | |
| CN113139594B (en) | Self-adaptive detection method for airborne image unmanned aerial vehicle target | |
| CN112881412B (en) | Method for detecting non-metal foreign matters in scrap steel products | |
| CN111738114B (en) | Vehicle target detection method based on anchor-free accurate sampling remote sensing image | |
| CN110991359A (en) | Satellite image target detection method based on multi-scale depth convolution neural network | |
| CN114943689A (en) | Method for detecting components of steel cold-rolling annealing furnace based on semi-supervised learning | |
| CN111178438A (en) | ResNet 101-based weather type identification method | |
| CN110599459A (en) | Underground pipe network risk assessment cloud system based on deep learning | |
| CN111091551A (en) | Method for detecting loss fault of brake beam strut opening pin of railway wagon | |
| CN114743102A (en) | Furniture board oriented flaw detection method, system and device | |
| CN111611956A (en) | Subway visual image-oriented track detection method and system | |
| CN115063725A (en) | Airplane skin defect identification system based on multi-scale self-adaptive SSD algorithm | |
| CN111241905A (en) | Power transmission line nest detection method based on improved SSD algorithm | |
| CN113516652A (en) | Battery surface defect and adhesive detection method, device, medium and electronic equipment | |
| CN117372854A (en) | Real-time detection method for hidden danger diseases of deep water structure of dam | |
| CN112085001A (en) | Tunnel recognition model and method based on multi-scale edge feature detection | |
| CN112949688A (en) | Motor train unit bottom plate rubber damage fault detection method, system and device | |
| CN115239733B (en) | Crack detection method and apparatus, terminal device and storage medium | |
| CN111539931A (en) | Appearance abnormity detection method based on convolutional neural network and boundary limit optimization | |
| CN116721291A (en) | Metal surface defect detection method based on improved YOLOv7 model |
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: 20210611 |