CN110310262A - A kind of method, apparatus and system for detection wheel tyre defect - Google Patents
A kind of method, apparatus and system for detection wheel tyre defect Download PDFInfo
- Publication number
- CN110310262A CN110310262A CN201910531159.9A CN201910531159A CN110310262A CN 110310262 A CN110310262 A CN 110310262A CN 201910531159 A CN201910531159 A CN 201910531159A CN 110310262 A CN110310262 A CN 110310262A
- Authority
- CN
- China
- Prior art keywords
- image
- mask
- defect
- tire
- segmentation
- 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
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
- G06T7/0004—Industrial image inspection
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T7/00—Image analysis
- G06T7/10—Segmentation; Edge 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
- G06T2207/10—Image acquisition modality
- G06T2207/10116—X-ray image
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T2207/00—Indexing scheme for image analysis or image enhancement
- G06T2207/20—Special algorithmic details
- G06T2207/20081—Training; Learning
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T2207/00—Indexing scheme for image analysis or image enhancement
- G06T2207/20—Special algorithmic details
- G06T2207/20084—Artificial neural networks [ANN]
Landscapes
- Engineering & Computer Science (AREA)
- Computer Vision & Pattern Recognition (AREA)
- Physics & Mathematics (AREA)
- General Physics & Mathematics (AREA)
- Theoretical Computer Science (AREA)
- Quality & Reliability (AREA)
- Image Analysis (AREA)
Abstract
The present invention relates to a kind of method, apparatus and system for detection wheel tyre defect.Wherein, method is the following steps are included: obtain the radioscopic image of a tire；Radioscopic image is inputted into trained Mask R-CNN model and carries out defect recognition；Output has the identification figure of flaw labeling；The Mask R-CNN model training process includes: the radioscopic image for obtaining tire, carries out the segmentation of setting pixel size using image segmentation software to every image；The image after segmentation is marked by visual pattern interpreter；Adaptive training is carried out to Mask R-CNN model using the image after label as training set.Compared with prior art, the present invention can carry out defects detection and classification simultaneously to detected image by trained Mask R-CNN model, so that the inspection accuracy to defect significantly improves.
Description
Technical field
The present invention relates to image identification technical fields, more particularly, to a kind of method, apparatus for detection wheel tyre defect
And system.
Background technique
With the fast development of automobile industry, the usage amount of tire is continuously increased, and tire sole mass just becomes people's wealth
The guarantee with life security is produced, so just becoming the pith of tire production to the quality testing of each factory tire.It passes
The software of the tire detecting system of system is complicated for operation, the practicability is poor, and most important defects detection part is by artificial observation wheel
The X-ray of tire acquires figure, this greatly reduces the correctness and efficiency of tire checking.Using computer to x-ray image into
Row analysis and identification, can greatly improve working efficiency, effectively overcome the erroneous judgement as caused by artificial origin in artificial evaluation
With fail to judge, so that Appraisal process is objectified, it is scientific and standardization.Establish and improve X-ray tire defect inspection and analysis
Processing system has very important significance to promotion tire quality and automotive safety aspect.
Currently used tire defects detection algorithm is to realize that wheel tyre defect detects automatically using x-ray imaging technology
's.But existing detection method has the following problems mostly: 1, needing to be arranged according to different types of defect characteristic suitable
The threshold value or parameter of specified defect, so that these inside tires defects detections and recognizer are in practical applications by certain journey
The limitation of degree, without universality.2, for the tire radioscopic image containing internal flaw, due to the intrinsic knot of tire itself
Structure shading, background shading are easy to produce aliasing with defect texture and are difficult to distinguish, so being not easy to judge complicated image defect
In the presence of, and defect kind, shape and the variation on boundary can all make the detection algorithm robustness dependent on geometrical characteristic poor, be easy
Generate missing inspection erroneous detection；Or big, defect extracts defect geometry shape that is imperfect and extracting to defect area size estimation deviation
Shape difference is big.
Summary of the invention
It is an object of the present invention to overcome the above-mentioned drawbacks of the prior art and provide one kind for detecting tire
The method, apparatus and system of defect.
The purpose of the present invention can be achieved through the following technical solutions:
A method of for detection wheel tyre defect, comprising the following steps:
S1, the radioscopic image for obtaining a tire；
S2, radioscopic image is inputted into trained Mask R-CNN model progress defect recognition；
S3, output have the identification figure of flaw labeling；
The Mask R-CNN model training process includes:
A1, the radioscopic image for obtaining tire carry out setting pixel size using image segmentation software to every image and divide
It cuts；
A2, the image after segmentation is marked by visual pattern interpreter；
A3, adaptive training is carried out to Mask R-CNN model using the image after label as training set.
Further, in the step A1, image segmentation size is 1024*1024 pixel.
Further, in the step A1, the lap of image segmentation is 150*150 pixel.
Further, in the step A2, no less than 3000 images are gone out to every a kind of flaw labeling；Step A3
In, the sample size of training set is no less than 10000.
Further, it is specific as follows to carry out adaptive training process for the Mask R-CNN model:
B1, training set is input in the good neural network of pre-training and obtains corresponding characteristic pattern；
B2, a area-of-interest is made a reservation for the setting of each of characteristic pattern point, so that it is interested to obtain multiple candidates
Region；
B3, the full convolutional network progress two-value classification of candidate interest region feeding depth and frame recurrence are filtered；
B4, ROI Align operation is carried out to filtered candidate area-of-interest, i.e., first by the pixel of original image and characteristic pattern
Value is mapped, and characteristic pattern and fixed feature are mapped, area-of-interest is marked；
B5, the characteristic pattern progress size by ROI Align operation is fixed, that is, traverses each candidate region, keeps
Floating number boundary, which is not done, to be quantified, and candidate region is divided into N*N unit, the boundary of each unit, which is not also done, to be quantified, each
It is calculated in unit and fixes four coordinate positions, the value of four positions is calculated with the method for bilinear interpolation, carry out maximum pond
Operation；
B6, the tire image that dimension and size have been unified is input to box mark module and categorization module is trained,
Output wheel tyre defect detection model.
Further, the training loss function of the Mask R-CNN model are as follows:
Lfinal=L ({ pi},{ti})+(Lcls+Lbox+Lmask)
In formula, LfinalIndicate training loss function, LclsIndicate the Classification Loss value of frame, LboxIndicate the recurrence damage of frame
Mistake value, LmaskIndicate the penalty values of the part Mask.
A kind of device for detection wheel tyre defect, the device include processor and memory, the processor
The data in memory are called to execute program, for realizing any above-mentioned method for detection wheel tyre defect.
A kind of system for detection wheel tyre defect, comprising:
Input unit, for obtaining the radioscopic image of a tire；
Recognition unit carries out defect recognition for radioscopic image to be inputted trained Mask R-CNN model；
Output unit, for exporting the identification figure for having flaw labeling；
The Mask R-CNN model training process includes:
A1, the radioscopic image for obtaining tire carry out setting pixel size using image segmentation software to every image and divide
It cuts；
A2, the image after segmentation is marked by visual pattern interpreter；
A3, adaptive training is carried out to Mask R-CNN model using the image after label as training set.
Compared with prior art, the invention has the following advantages that
1, the present invention proposes a kind of high-precision tire defect inspection method, by trained Mask R-CNN model into
The intelligent measurement of row wheel tyre defect, can rapidly and efficiently detect wheel tyre defect.Trained Mask R-CNN model can
Defects detection and classification are carried out simultaneously to detected image, so that the inspection accuracy to defect significantly improves.
2, the present invention carries out the segmentation of setting pixel size by image segmentation software to original image, and passes through vision
The image after segmentation is marked in annotation of images device, enables and obtains sufficient amount of trained sample in Mask R-CNN model
This, is avoided overfitting problem.In existing deep learning, whether the amount of training set is enough, has very directly to training result
Influence.Meanwhile pixel size segmentation is carried out to image, the accuracy of training result can be further improved.
Detailed description of the invention
Fig. 1 is flow diagram of the invention.
Fig. 2 is the training flow diagram of Mask R-CNN model.
Fig. 3 is the flow diagram that Mask R-CNN model carries out adaptive training.
Specific embodiment
The present invention is described in detail with specific embodiment below in conjunction with the accompanying drawings.The present embodiment is with technical solution of the present invention
Premised on implemented, the detailed implementation method and specific operation process are given, but protection scope of the present invention is not limited to
Following embodiments.
As described in Figure 1, a kind of method for detection wheel tyre defect is present embodiments provided, comprising the following steps:
Step S1, the radioscopic image of a tire is obtained；
Step S2, radioscopic image is inputted into trained Mask R-CNN model and carries out defect recognition；
Step S3, output has the identification figure of flaw labeling.
This method can fast and accurately detect that bubble, connector are opened in sidewall and tire, split seam, dilute line, sundries, dilute line,
Steel wire bent, cord cross-lapping, cord disconnect, are pressed into the problems such as impurity, bead distortion, before independent of artificial detection
It puts, defect tire can be gone out with accurate detection by this method.
In step s 2, Mask R-CNN model training process includes:
Step A1, the original X-rays image that a large amount of tire is obtained from database is soft using image segmentation to every image
Part carries out the segmentation of setting pixel size.
This example is split tire image using splitimage image segmentation software, because of the pixel of tire image
Substantially 1500*10000 or so, image are not suitable for greatly very much doing target identification, so such image is cut into 1024*1024
Size pixel, the lap of image segmentation are 150*150 pixel.In the task based on deep learning, sufficient amount of instruction
Practicing sample can be avoided serious overfitting problem.The meaning of such segmentation can mainly be divided in limited training set
More images out, so that the picture number in training set is greatly improved, because in existing deep learning, the amount of training set
Whether enough, there is very direct influence to training result.Meanwhile pixel size is carried out to image and divides and can further mention
The accuracy of high training result.
Step A2, the image after segmentation is marked by visual pattern interpreter.
The present embodiment is using visual pattern interpreter (Visual Geometry Group, hereinafter referred to as VGG) to tire figure
As being marked, ROI region, such as sundries defect are marked, marks 3000 images altogether.Other defects are lacked with sundries
It falls into, 3000 images is gone out to every a kind of flaw labeling respectively, centre will appear in an image that there are multiple defects.Due to VGG
The image of generation can change into coordinate form, and Mask R-CNN can only read coordinate, it is possible to be applied directly to Mask R-
In CNN, without additional conversion.
Step A3, mark 10000 or so image is put into Mask R-CNN model as training set and is carried out certainly
Adaptation training.
As shown in Fig. 2, Mask R-CNN model progress adaptive training process is specific as follows:
Step B1, training set is input in the good neural network of pre-training, it is corresponding obtains tire image different phase
Feature map (characteristic pattern).R-FPN (depth residual error network) neural network, Web vector graphic tool have been used in the present embodiment
There is the framework from top to bottom of lateral connection, inputs building from single scale and net interior feature pyramid.Tire can preferably be obtained
The characteristic pattern of image.
Step B2, a ROI (area-of-interest) is made a reservation for each of feature map point setting, to obtain
Multiple candidate ROI.
The effect of this step is because in order to improve pace of learning and efficiency being all only to close in the traditional R-CNN network architecture
ROI is infused, and individually assesses convolutional network in each ROI, is trained and predicts, so needing to select suitable ROI.For example it selects
The rejected region for selecting out tire image, with rectangle candidate frame iris out come, it is possible that mark is not the position of defect, because
This, the present embodiment further carries out step B3.
Step B3, candidate ROI is sent into (the full reel wire network product of depth) the progress two-value classification of RPN network and frame is returned and carried out
Filtering, the new ROI of generation are exactly a part of Anchor frame work that RPN chooses from each pixel Anchor frame generated
For ROI.
The step filters out some meaningless Anchor frames by RPN, each pixel can give birth to above tire image
At an Anchor frame, frame can be very more, are unfavorable for selecting ROI, so some frames are filtered out using RPN, so that leaving one
Frame segment is more advantageous to the selection and next Align operation of ROI.For example filter out some labels mistake of tire image
ROI region so that the ROI region stayed is all relatively accurate.
Step B4, ROI Align operation is carried out to filtered ROI, i.e., it is first that the pixel value of original image and characteristic pattern is corresponding
Get up, then characteristic pattern and fixed feature are mapped, accurately ROI region can be marked.
Since from the ROI on the ROI to characteristic pattern on input figure, ROI Pooling is directly to pass through round
It is obtaining as a result, however the value that is directly taken with round, the output that ROI Pooling can be made to obtain later may on original image
ROI on not.The effect of ROI Align is exactly mainly to eliminate the floor operation of ROI Pooling, and to be each
The ROI region in original image can be better aligned in the feature that ROI is obtained.Such as in tire checking, one determined by step B3
Part ROI region candidate frame may not be aligned in well in tire original image, some may enclose the edge of defect, have
Tire original image may also can be better aligned from defect mesosphere mistake, ROI Align a bit, so that ROI region candidate frame circle
It is more accurate.
Step B5, size will be carried out by the characteristic pattern of ROI Align operation to fix, that is, traverse each candidate region,
It keeps floating number boundary not do to quantify.Candidate region is divided into N*N unit, the N in the present embodiment takes 7, each unit
Boundary, which is not also done, to be quantified.In each cell calculate fix four coordinate positions, with the method for bilinear interpolation calculate this four
Then the value of a position carries out maximum pondization operation.
It is in different size by the step B4 ROI exported, however the box label and classification in subsequent step are all
The image dimension size for needing to input must be consistent, so ensure that tire image dimension is consistent with size by the step
Property.
Step B6, the tire image that dimension and size have been unified is input to box mark module (to tire rejected region
Carry out box label) and categorization module (to wheel tyre defect classify) be trained, at this moment can export a wheel tyre defect
Detection model.Classification can carry out simultaneously with box label, make the more accurate of the last model inspection trained.
The training loss function of Mask R-CNN model are as follows:
Lfinal=L ({ pi},{ti})+(Lcls+Lbox+Lmask)
In formula, LfinalIndicate training loss function, LclsIndicate the Classification Loss value of frame, LboxIndicate the recurrence damage of frame
Mistake value, LmaskIndicate the penalty values of the part Mask.
The preferred embodiment of the present invention has been described in detail above.It should be appreciated that those skilled in the art without
It needs creative work according to the present invention can conceive and makes many modifications and variations.Therefore, all technologies in the art
Personnel are available by logical analysis, reasoning, or a limited experiment on the basis of existing technology under this invention's idea
Technical solution, all should be within the scope of protection determined by the claims.
Claims (8)
1. a kind of method for detection wheel tyre defect, which comprises the following steps:
S1, the radioscopic image for obtaining a tire；
S2, radioscopic image is inputted into trained Mask R-CNN model progress defect recognition；
S3, output have the identification figure of flaw labeling；
The Mask R-CNN model training process includes:
A1, the radioscopic image for obtaining tire, the segmentation of setting pixel size is carried out to every image using image segmentation software；
A2, the image after segmentation is marked by visual pattern interpreter；
A3, adaptive training is carried out to Mask R-CNN model using the image after label as training set.
2. the method according to claim 1 for detection wheel tyre defect, which is characterized in that in the step A1, figure
As segmentation size is 1024*1024 pixel.
3. the method according to claim 2 for detection wheel tyre defect, which is characterized in that in the step A1, figure
As the lap of segmentation is 150*150 pixel.
4. the method according to claim 1 for detection wheel tyre defect, which is characterized in that right in the step A2
Every one kind flaw labeling goes out no less than 3000 images；In step A3, the sample size of training set is no less than 10000.
5. the method according to claim 1 for detection wheel tyre defect, which is characterized in that the Mask R-CNN mould
It is specific as follows that type carries out adaptive training process:
B1, training set is input in the good neural network of pre-training and obtains corresponding characteristic pattern；
B2, a area-of-interest is made a reservation for the setting of each of characteristic pattern point, to obtain multiple candidate area-of-interests；
B3, the full convolutional network progress two-value classification of candidate interest region feeding depth and frame recurrence are filtered；
B4, ROI Align operation is carried out to filtered candidate area-of-interest, i.e., first by the pixel value pair of original image and characteristic pattern
It should get up, characteristic pattern and fixed feature are mapped, area-of-interest is marked；
B5, the characteristic pattern progress size by ROI Align operation is fixed, that is, traverses each candidate region, keeps floating-point
Number boundary, which is not done, to be quantified, and candidate region is divided into N*N unit, the boundary of each unit, which is not also done, to be quantified, in each unit
Four coordinate positions are fixed in middle calculating, and the value of four positions is calculated with the method for bilinear interpolation, carry out maximum pondization operation；
B6, the tire image that dimension and size have been unified is input to box mark module and categorization module is trained, output
Tire defects detection model.
6. the method according to claim 1 for detection wheel tyre defect, which is characterized in that the Mask R-CNN mould
The training loss function of type are as follows:
Lfinal=L ({ pi},{ti})+(Lcls+Lbox+Lmask)
In formula, LfinalIndicate training loss function, LclsIndicate the Classification Loss value of frame, LboxIndicate the recurrence loss of frame
Value, LmaskIndicate the penalty values of the part Mask.
7. a kind of device for detection wheel tyre defect, the device includes processor and memory, which is characterized in that institute
Stating processor calls the data in memory to execute program, is used for detection wheel for realizing as described in claim 1~6 is any
The method of tyre defect.
8. a kind of system for detection wheel tyre defect characterized by comprising
Input unit, for obtaining the radioscopic image of a tire；
Recognition unit carries out defect recognition for radioscopic image to be inputted trained Mask R-CNN model；
Output unit, for exporting the identification figure for having flaw labeling；
The Mask R-CNN model training process includes:
A1, the radioscopic image for obtaining tire, the segmentation of setting pixel size is carried out to every image using image segmentation software；
A2, the image after segmentation is marked by visual pattern interpreter；
A3, adaptive training is carried out to Mask R-CNN model using the image after label as training set.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201910531159.9A CN110310262A (en) | 2019-06-19 | 2019-06-19 | A kind of method, apparatus and system for detection wheel tyre defect |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201910531159.9A CN110310262A (en) | 2019-06-19 | 2019-06-19 | A kind of method, apparatus and system for detection wheel tyre defect |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| CN110310262A true CN110310262A (en) | 2019-10-08 |
Family
ID=68077509
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201910531159.9A Pending CN110310262A (en) | 2019-06-19 | 2019-06-19 | A kind of method, apparatus and system for detection wheel tyre defect |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN110310262A (en) |
Cited By (15)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN110414499A (en) * | 2019-07-26 | 2019-11-05 | 第四范式（北京）技术有限公司 | Text position localization method and system and model training method and system |
| CN110826416A (en) * | 2019-10-11 | 2020-02-21 | 佛山科学技术学院 | Bathroom ceramic surface defect detection method and device based on deep learning |
| CN111079817A (en) * | 2019-12-12 | 2020-04-28 | 哈尔滨市科佳通用机电股份有限公司 | Method for identifying fault image of cross beam of railway wagon |
| CN111127416A (en) * | 2019-12-19 | 2020-05-08 | 武汉珈鹰智能科技有限公司 | Computer vision-based automatic detection method for surface defects of concrete structure |
| CN111161233A (en) * | 2019-12-25 | 2020-05-15 | 武汉科技大学 | Method and system for detecting defects of punched leather |
| CN111179239A (en) * | 2019-12-24 | 2020-05-19 | 浙江大学 | Tire X-ray flaw detection method for performing re-ranking by using background features |
| CN111191714A (en) * | 2019-12-28 | 2020-05-22 | 浙江大学 | Intelligent identification method for bridge appearance damage diseases |
| CN111260626A (en) * | 2020-01-15 | 2020-06-09 | 佛山科学技术学院 | Workpiece wear detection method and system based on deep learning |
| CN111402214A (en) * | 2020-03-07 | 2020-07-10 | 西南交通大学 | Neural network-based automatic detection method for breakage defect of catenary dropper current-carrying ring |
| CN112069907A (en) * | 2020-08-11 | 2020-12-11 | 盛视科技股份有限公司 | X-ray machine image recognition method, device and system based on example segmentation |
| CN112557063A (en) * | 2020-12-01 | 2021-03-26 | 安徽信息工程学院 | Vision-based tread detection system of straddle type single-track tire |
| CN112862811A (en) * | 2021-03-10 | 2021-05-28 | 华中科技大学 | Material microscopic image defect identification method, equipment and device based on deep learning |
| CN113850749A (en) * | 2020-06-09 | 2021-12-28 | 英业达科技有限公司 | Method for training defect detector |
| CN114036971A (en) * | 2021-11-18 | 2022-02-11 | 中科星睿科技(北京)有限公司 | Oil tank information generation method, oil tank information generation device, electronic device, and computer-readable medium |
| CN117115082A (en) * | 2023-07-12 | 2023-11-24 | 钛玛科（北京）工业科技有限公司 | Method and equipment for detecting overlap quality of tire |
Citations (6)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN108364006A (en) * | 2018-01-17 | 2018-08-03 | 超凡影像科技股份有限公司 | Medical Images Classification device and its construction method based on multi-mode deep learning |
| CN108985337A (en) * | 2018-06-20 | 2018-12-11 | 中科院广州电子技术有限公司 | A kind of product surface scratch detection method based on picture depth study |
| CN109300119A (en) * | 2018-09-11 | 2019-02-01 | 石家庄铁道大学 | Detection method, detection device and the terminal device in steel structure surface corrosion region |
| CN109472769A (en) * | 2018-09-26 | 2019-03-15 | 成都数之联科技有限公司 | A kind of bad image defect detection method and system |
| US20190154442A1 (en) * | 2016-01-15 | 2019-05-23 | Fugro Roadware Inc. | High speed stereoscopic pavement surface scanning system and method |
| CN109816669A (en) * | 2019-01-30 | 2019-05-28 | 云南电网有限责任公司电力科学研究院 | A kind of improvement Mask R-CNN image instance dividing method identifying power equipments defect |
-
2019
- 2019-06-19 CN CN201910531159.9A patent/CN110310262A/en active Pending
Patent Citations (6)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US20190154442A1 (en) * | 2016-01-15 | 2019-05-23 | Fugro Roadware Inc. | High speed stereoscopic pavement surface scanning system and method |
| CN108364006A (en) * | 2018-01-17 | 2018-08-03 | 超凡影像科技股份有限公司 | Medical Images Classification device and its construction method based on multi-mode deep learning |
| CN108985337A (en) * | 2018-06-20 | 2018-12-11 | 中科院广州电子技术有限公司 | A kind of product surface scratch detection method based on picture depth study |
| CN109300119A (en) * | 2018-09-11 | 2019-02-01 | 石家庄铁道大学 | Detection method, detection device and the terminal device in steel structure surface corrosion region |
| CN109472769A (en) * | 2018-09-26 | 2019-03-15 | 成都数之联科技有限公司 | A kind of bad image defect detection method and system |
| CN109816669A (en) * | 2019-01-30 | 2019-05-28 | 云南电网有限责任公司电力科学研究院 | A kind of improvement Mask R-CNN image instance dividing method identifying power equipments defect |
Non-Patent Citations (4)
| Title |
|---|
| HEIHEIYA: "【Deep Learning】深度学习图像标注工具VGG Image Annotator (VIA)使用教程", 《HTTPS://BLOG.CSDN.NET/HEIHEIYA/ARTICLE/DETAILS/81530952》 * |
| I_LINDA: "Mask RCNN_ Mask RCNN详细流程解析", 《HTTPS://BLOG.CSDN.NET/QQ_34713831/ARTICLE/DETAILS/85123905》 * |
| MAX FERGUSON 等: "Detection and Segmentation of Manufacturing Defects with Convolutional Neural Networks and Transfer Learning", 《COMPUTER VISION AND PATTERN RECOGNITION》 * |
| 马可菠萝: "SplitImage 2.2 图像拆分工具", 《HTTPS://WWW.MACBL.COM/PLUGIN/ADOBE-AFTER-EFFECTS/DISTORTION/SPLITIMAGE》 * |
Cited By (21)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN110414499B (en) * | 2019-07-26 | 2021-06-04 | 第四范式（北京）技术有限公司 | Text position positioning method and system and model training method and system |
| CN110414499A (en) * | 2019-07-26 | 2019-11-05 | 第四范式（北京）技术有限公司 | Text position localization method and system and model training method and system |
| CN110826416A (en) * | 2019-10-11 | 2020-02-21 | 佛山科学技术学院 | Bathroom ceramic surface defect detection method and device based on deep learning |
| CN110826416B (en) * | 2019-10-11 | 2023-05-30 | 佛山科学技术学院 | Bathroom ceramic surface defect detection method and device based on deep learning |
| CN111079817B (en) * | 2019-12-12 | 2020-11-27 | 哈尔滨市科佳通用机电股份有限公司 | Method for identifying fault image of cross beam of railway wagon |
| CN111079817A (en) * | 2019-12-12 | 2020-04-28 | 哈尔滨市科佳通用机电股份有限公司 | Method for identifying fault image of cross beam of railway wagon |
| CN111127416A (en) * | 2019-12-19 | 2020-05-08 | 武汉珈鹰智能科技有限公司 | Computer vision-based automatic detection method for surface defects of concrete structure |
| CN111179239A (en) * | 2019-12-24 | 2020-05-19 | 浙江大学 | Tire X-ray flaw detection method for performing re-ranking by using background features |
| CN111179239B (en) * | 2019-12-24 | 2022-04-26 | 浙江大学 | Tire X-ray flaw detection method for performing re-ranking by using background features |
| CN111161233A (en) * | 2019-12-25 | 2020-05-15 | 武汉科技大学 | Method and system for detecting defects of punched leather |
| CN111191714A (en) * | 2019-12-28 | 2020-05-22 | 浙江大学 | Intelligent identification method for bridge appearance damage diseases |
| CN111260626A (en) * | 2020-01-15 | 2020-06-09 | 佛山科学技术学院 | Workpiece wear detection method and system based on deep learning |
| CN111402214A (en) * | 2020-03-07 | 2020-07-10 | 西南交通大学 | Neural network-based automatic detection method for breakage defect of catenary dropper current-carrying ring |
| CN113850749A (en) * | 2020-06-09 | 2021-12-28 | 英业达科技有限公司 | Method for training defect detector |
| CN112069907A (en) * | 2020-08-11 | 2020-12-11 | 盛视科技股份有限公司 | X-ray machine image recognition method, device and system based on example segmentation |
| CN112557063A (en) * | 2020-12-01 | 2021-03-26 | 安徽信息工程学院 | Vision-based tread detection system of straddle type single-track tire |
| CN112862811A (en) * | 2021-03-10 | 2021-05-28 | 华中科技大学 | Material microscopic image defect identification method, equipment and device based on deep learning |
| CN114036971A (en) * | 2021-11-18 | 2022-02-11 | 中科星睿科技(北京)有限公司 | Oil tank information generation method, oil tank information generation device, electronic device, and computer-readable medium |
| CN114036971B (en) * | 2021-11-18 | 2022-05-17 | 中科星睿科技(北京)有限公司 | Oil tank information generation method, oil tank information generation device, electronic device, and computer-readable medium |
| CN117115082A (en) * | 2023-07-12 | 2023-11-24 | 钛玛科（北京）工业科技有限公司 | Method and equipment for detecting overlap quality of tire |
| CN117115082B (en) * | 2023-07-12 | 2024-04-05 | 钛玛科（北京）工业科技有限公司 | Method and equipment for detecting overlap quality of tire |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN110310262A (en) | A kind of method, apparatus and system for detection wheel tyre defect | |
| Su et al. | Deep learning-based solar-cell manufacturing defect detection with complementary attention network | |
| CN110136101B (en) | Tire X-ray defect detection method based on twinning distance comparison | |
| CN103871053B (en) | Vision conspicuousness-based cloth flaw detection method | |
| CN108711148B (en) | Tire defect intelligent detection method based on deep learning | |
| CN113920107A (en) | Insulator damage detection method based on improved yolov5 algorithm | |
| CN113643268A (en) | Industrial product defect quality inspection method and device based on deep learning and storage medium | |
| CN111932511B (en) | Electronic component quality detection method and system based on deep learning | |
| CN111382785A (en) | GAN network model and method for realizing automatic cleaning and auxiliary marking of sample | |
| CN112329588A (en) | Pipeline fault detection method based on fast R-CNN | |
| US20220076404A1 (en) | Defect management apparatus, method and non-transitory computer readable medium | |
| CN111476307A (en) | Lithium battery surface defect detection method based on depth field adaptation | |
| CN111753877B (en) | Product quality detection method based on deep neural network migration learning | |
| CN113420619A (en) | Remote sensing image building extraction method | |
| CN115294033A (en) | Tire belt layer difference level and misalignment defect detection method based on semantic segmentation network | |
| Fang et al. | Automatic zipper tape defect detection using two-stage multi-scale convolutional networks | |
| CN114565594A (en) | Image anomaly detection method based on soft mask contrast loss | |
| CN109410241A (en) | The metamorphic testing method of image-region growth algorithm | |
| CN109447968A (en) | The metamorphic testing system of image-region growth algorithm | |
| CN114078106A (en) | Defect detection method based on improved Faster R-CNN | |
| CN113191352A (en) | Water meter pointer reading identification method based on target detection and binary image detection | |
| Li et al. | Semantic-Segmentation-Based rail fastener state recognition algorithm | |
| CN113642662B (en) | Classification detection method and device based on lightweight classification model | |
| Wang et al. | Defect transformer: An efficient hybrid transformer architecture for surface defect detection | |
| CN114742803A (en) | Platelet aggregation detection method combining deep learning and digital image processing algorithm |
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: 20191008 |