CN109816669A - A kind of improvement Mask R-CNN image instance dividing method identifying power equipments defect - Google Patents
A kind of improvement Mask R-CNN image instance dividing method identifying power equipments defect Download PDFInfo
- Publication number
- CN109816669A CN109816669A CN201910089308.0A CN201910089308A CN109816669A CN 109816669 A CN109816669 A CN 109816669A CN 201910089308 A CN201910089308 A CN 201910089308A CN 109816669 A CN109816669 A CN 109816669A
- Authority
- CN
- China
- Prior art keywords
- power equipments
- region
- equipments defect
- mask
- refinement
- 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.)
- Withdrawn
Links
Abstract
This application discloses a kind of improvement Mask R-CNN image instance dividing methods for identifying power equipments defect, comprising: building convolutional neural networks；It reads in power equipments defect pretreatment picture power equipments defect pretreatment picture is input in convolutional neural networks, convolutional neural networks carry out feature extraction to power equipments defect pretreatment picture, obtain the region containing feature；The region containing feature is refined by RPN network, in region after refinement, RoIAlign operation is carried out by bilinearity difference approach to handle the region after refinement, each fixed-size characteristic pattern of interest Area generation, pass through classification, coordinate information, and mask information, obtain the segmentation of power equipments defect example.By being improved to Mask R-CNN image instance dividing method, retain the basic effect of Mask R-CNN segmentation, bilinear interpolation speed during raising RoIAlign, keep mapping process abundant simultaneously, power equipments defect pretreatment picture all pixels are uniformly utilized, keep segmentation effect more obvious.
Description
Technical field
This application involves image object detection and segmentation technologies more particularly to a kind of identification power equipments defect to change
Into Mask R-CNN image instance dividing method.
Background technique
With the rapid development of power grid, what electrification also developed is getting faster, in order to guarantee the normal operation of power equipment,
If closing power equipment during operation and carrying out coherence check to it, this is infeasible.Power equipment is some simultaneously
What hidden danger and defect were often hidden again, therefore can not find in time, it is even more impossible to exclude in time.
When there is eclipse phenomena between power equipment target, if simply using arest neighbors interpolation method (Nearest
Neighbor Interpolation), although the less speed of calculation amount is fast, spatial symmetry (Alignment) is in certain journey
It is destroyed on degree, that is, introduces image fault.In order to preferably promote testing result, image, semantic need to be used to divide (Image
Semantic Segmentation), the part for making multiple target mutually block overlapping is split from image, carries out Pixel-level
Other picture material is intensively classified, and image pixel is traversed, and is realized and is marked to the semantic information of all pixels.However, semantic segmentation
Certain class of prediction is only exported on the image as a result, not distinguishing to the specific object i.e. example of class, to reach to image
The purpose of power equipments defect target detection is blocked, the individual of class need to be distinguished, i.e., example divides (Instance
Segmentation), the image instance based on Mask R-CNN network is partitioned into mainstream, on the one hand which need to carry out pixel
The image of rank is intensively classified, and on the other hand need to also be distinguished under the premise of predicting classification to different instances.Due to Mask
R-CNN is to divide pixel-by-pixel, therefore when using bilinear interpolation algorithm, while image segmentation precision improves, will lead to meter
The disadvantages of speed reduces, splitting speed is slower is calculated, therefore limited as the practicability of power equipments defect detection system.
Summary of the invention
This application provides a kind of improvement Mask R-CNN image instance dividing methods for identifying power equipments defect, with solution
Certainly the deficiencies in the prior art.
This application provides a kind of improvement Mask R-CNN image instance dividing method for identifying power equipments defect, methods
The following steps are included:
Step S1: building convolutional neural networks；
Step S2: reading in power equipments defect and pre-process picture, and power equipments defect pretreatment picture is input to convolution
In neural network, convolutional neural networks carry out feature extraction to power equipments defect pretreatment picture, obtain the area containing feature
Domain；
Step S3: refining the region containing feature by RPN (Region Proposal Network) network,
Realize the recurrence to increment, the region after being refined；
Step S4: in the region after refinement, after carrying out RoIAlign operation to refinement by bilinearity difference approach
Region is handled, and each fixed-size characteristic pattern of interest Area generation is made；
Step S5: according to each fixed-size characteristic pattern of interest Area generation, the class obtained by Faster-RCNN
Not, coordinate information, and the mask information that full convolutional network obtains, obtain the segmentation of power equipments defect example.
It is selectable, in the region after refinement, after carrying out RoIAlign operation to refinement by bilinearity difference approach
Region handled, make each fixed-size characteristic pattern of interest Area generation include:
Step S41: in the region after refinement, being rounded floating number, takes after fractional part is carried out multiplying power amplification
Whole carry out operation, operation result pass through shift right operation again and obtain true fractional part, in addition the operation knot of integer part
Fruit obtains interpolation result to the end；
Step S42: RoIAlign operation is carried out by difference result, the region after refinement is handled, make each interest
The fixed-size characteristic pattern of Area generation.
Selectable, being carried out by bilinearity difference approach will when RoIAlign operation handles the region after refinement
Power equipments defect pre-processes the geometric center of picture and the geometric center alignment of target image.
It is selectable, according to each fixed-size characteristic pattern of interest Area generation, the class obtained by Faster-RCNN
Not, coordinate information, and the mask information that full convolutional network obtains, before obtaining the segmentation of power equipments defect example further include:
Classification information and coordinate information are obtained by Faster-RCNN；
Mask information is obtained by full convolutional network
From the above technical scheme, this application provides a kind of improvement Mask R-CNN figures for identifying power equipments defect
As example dividing method, comprising: building convolutional neural networks；It reads in power equipments defect and pre-processes picture, power equipment is lacked
It falls into pretreatment picture to be input in convolutional neural networks, convolutional neural networks carry out feature to power equipments defect pretreatment picture
It extracts, obtains the region containing feature；The region containing feature is refined by RPN network, realizes the recurrence to increment,
Region after being refined；In region after refinement, after carrying out RoIAlign operation to refinement by bilinearity difference approach
Region handled, make each fixed-size characteristic pattern of interest Area generation；According to each interest Area generation fixed ruler
Very little characteristic pattern, the classification obtained by Faster-RCNN, coordinate information, and the mask information that full convolutional network obtains, obtain
Divide to power equipments defect example.The application accelerates the ROIAlign of network by improving to bilinear interpolation algorithm
Process, as far as possible reduction interpolation quantity calculation, first by pair of power equipments defect pretreatment picture and target image geometric center
Together, interpolating pixel quantity is needed by the reduction of center alignment thereof, the floating-point operation of Interpolation Process is then converted into shift operation,
Reduce execution cycle.Mask R-CNN image instance dividing method is improved by the above method, improved method exists
It completes under conditions of splitting power equipments defect image and background information, had both remained the segmentation of Mask R-CNN
Basic effect, and the speed of bilinear interpolation during RoIAlign is significantly improved, while keeping mapping process sufficiently and equal
The even all pixels that power equipments defect pretreatment picture is utilized, so that segmentation effect is more obvious, and splitting speed
More rapidly.
Detailed description of the invention
In order to illustrate more clearly of the technical solution of the application, letter will be made to attached drawing needed in the embodiment below
Singly introduce, it should be apparent that, for those of ordinary skills, without any creative labor,
It is also possible to obtain other drawings based on these drawings.
Fig. 1 is a kind of improvement Mask R-CNN image instance dividing method for identifying power equipments defect provided by the present application
The flow chart of one embodiment；
Fig. 2 is a kind of improvement Mask R-CNN image instance dividing method for identifying power equipments defect provided by the present application
The flow chart of another embodiment；
Fig. 3 is a kind of improvement Mask R-CNN image instance dividing method for identifying power equipments defect provided by the present application
Work step schematic diagram；
Fig. 4 is a kind of improvement Mask R-CNN image instance dividing method for identifying power equipments defect provided by the present application
Target image and source image pixels between mapping relations figure；
Fig. 5 is a kind of improvement Mask R-CNN image instance dividing method for identifying power equipments defect provided by the present application
Improved method after mapping relations figure between source images and target image；
Fig. 6 is that RoIAlign optimizes speeding scheme block diagram.
Specific embodiment
Below with reference to the attached drawing in the application, the technical scheme in the embodiment of the application is clearly and completely described,
Obviously, described embodiment is only a part of the embodiment of the application, instead of all the embodiments.Based in the application
Embodiment, every other embodiment obtained by those of ordinary skill in the art without making creative efforts,
It shall fall within the protection scope of the present invention.
Many details are explained in the following description in order to fully understand the application, but the application can be with
It is different from the other modes that describe again using other to implement, those skilled in the art can be without prejudice to the application intension
In the case of do similar popularization, therefore the application is not limited by the specific embodiments disclosed below.
Referring to Fig. 1, for a kind of improvement Mask R-CNN image instance for identifying power equipments defect point provided by the present application
The flow chart of one embodiment of segmentation method.This application provides a kind of improvement Mask R-CNN image for identifying power equipments defect is real
Example dividing method, method the following steps are included:
Step S101: building convolutional neural networks；
Step S102: reading in power equipments defect and pre-process picture, by power equipments defect pretreatment picture input
Into convolutional neural networks, the convolutional neural networks carry out feature extraction to power equipments defect pretreatment picture, obtain
To the region containing feature；
Step S103: the region containing feature is carried out by RPN (Region Proposal Network) network
The recurrence to increment, the region after being refined are realized in refinement；
Step S104: in the region after refinement, after carrying out RoIAlign operation to refinement by bilinearity difference approach
Region handled, make each fixed-size characteristic pattern of interest Area generation；
Step S105: obtaining classification information and coordinate information by Faster-RCNN, obtains exposure mask by full convolutional network
Information；
Step S106: it according to each fixed-size characteristic pattern of interest Area generation, is obtained by Faster-RCNN
Classification, coordinate information, and the mask information that full convolutional network obtains obtains the segmentation of power equipments defect example.
It referring to fig. 2, is a kind of improvement Mask R-CNN image instance for identifying power equipments defect point provided by the present application
The flow chart of one embodiment of segmentation method.Referring to Fig. 3, for a kind of improvement MaskR- for identifying power equipments defect provided by the present application
The work step schematic diagram of CNN image instance dividing method.
This application provides a kind of improvement Mask R-CNN image instance dividing method for identifying power equipments defect, methods
The following steps are included:
Step S201: building convolutional neural networks；
Step S202: reading in power equipments defect and pre-process picture, by power equipments defect pretreatment picture input
Into the convolutional neural networks, the convolutional neural networks carry out feature to power equipments defect pretreatment picture and mention
It takes, obtains the region containing feature；
Step S203: the region containing feature is carried out by RPN (Region Proposal Network) network
The recurrence to increment, the region after being refined are realized in refinement；
Step S204: in the region after refinement, being rounded floating number, takes after fractional part is carried out multiplying power amplification
Whole carry out operation, operation result pass through shift right operation again and obtain true fractional part, in addition the operation knot of integer part
Fruit obtains interpolation result to the end；
Step S205: RoIAlign operation is carried out by difference result, the region after refinement is handled, made each emerging
The interesting fixed-size characteristic pattern of Area generation；
Step S206: obtaining classification information and coordinate information by Faster-RCNN, obtains exposure mask by full convolutional network
Information；
Step S207: it according to each fixed-size characteristic pattern of interest Area generation, is obtained by Faster-RCNN
Classification, coordinate information, and the mask information that full convolutional network obtains obtains the segmentation of power equipments defect example.
Selectable, being carried out by bilinearity difference approach will when RoIAlign operation handles the region after refinement
Power equipments defect pre-processes the geometric center of picture and the geometric center alignment of target image.
When RoIAlign does bilinear interpolation, source images and target image geometric center are aligned first, source figure
Picture as power equipments defect pre-processes picture, the specific implementation process is as follows:
When interest region executes RoIAlign step in characteristic pattern, for the processing speed for accelerating bilinear interpolation algorithm
Degree, should reduce interpolation quantity calculation as far as possible, and can be reduced by center alignment thereof needs interpolating pixel quantity.
In former bilinear interpolation algorithm, it is assumed that source images size m × n, target image size are a × b, then two images
Width and height ratio be respectively m/a and n/b, then certain pixel (dstX, dstY) in target image, can pass through interpolation ratio
Example maps back respective pixel location in source images (srcX, srcY), calculation formula are as follows:
Assuming that source images size is 7 × 7, target image size is 3 × 3, and the selection upper left corner is coordinate origin, then source images
It is respectively (3,3) and (1,1) with the respective center point coordinate of target image, it, should as far as possible when carrying out interpolation mapping
The even Pixel Information for using source images.Such as lower right corner pixel (2,2) in target image, using the calculating pair of former interpolation algorithm
Answering source images coordinate is srcX=2 (7/3)=4.67, chooses four apart from nearest pixel (4,4), (4,5), (5,4)
And (5,5) carry out bilinear interpolation operation, pixel (7,7) is unutilized in source images at this time, when to entire target image to
When source images map, the pixel utilized concentrates on the upper left side of image, and the pixel of lower right-most portion is unutilized, and with slotting
The increase of value ratio, non-uniform phenomenon is more obvious, such as Fig. 4 of the mapping between target image and source image pixels, in source images
Block of pixels without color is the non-uniform Distribution of unutilized pixel.
The alignment thereof for improving source images and target image, is improved to center point alignment for upper left angle alignment, calculates at this time
Formula is following formula:
Formula can be further deformed into:
The latter half of formula can be considered that the controlling elements of entire interpolation arithmetic, value can just be born, at this time source images
Coordinate center with target image is coordinate origin, and source images are reached with the center point coordinate of target image and are overlapped, exactly
Due to the presence of the controlling elements, so that mapping process is abundant and all pixels of source figure, mapping relations are uniformly utilized
Such as Fig. 5.
At this point, interpolation maps the more uniform pixel that source images are utilized, when avoiding characteristic pattern pond caused by region
Information is lost, so that target image interpolation is more preferable.
Floating-point operation is then converted into integer arithmetic, the specific implementation process is as follows:
Bilinear interpolation process will cause a large amount of floating-point operations, Mask R- due to the selection of different target picture size
There are numerous feature graph regions need to carry out RoIAlign operation in CNN network, needs a large amount of bilinear interpolation operation, floating-point fortune
Bilinear interpolation process more time-consuming one of reason when calculation.In view of integer arithmetic calculation amount is much smaller than floating-point operation, and move
Bit arithmetic computational efficiency is much higher than common multiplication and division operation, proposes that RoIAlign optimizes speeding scheme, step such as Fig. 6 based on this.
It is usually floating number that the pixel value f (x, y) at target image point (x, y), which is calculated, in bilinear interpolation, first right
The floating number is rounded, and is rounded after fractional part is carried out multiplying power amplification.Committed step is the selection of multiplying power, needs to integrate
Consider three factors.
It first has to guarantee certain precision, if multiplying power value is smaller, pixel value obtains decimal and do not amplified effectively, finally
Rounding can give up more numerical value, it is larger to may result in interpolation result error；Next times of rate score should not be too large, and multiplying power is excessive
It is likely to result in the spilling of calculating process numerical value, i.e., the maximum value that can be expressed beyond data type；Finally in order to using displacement
Operation is accelerated, and need to make multiplying power is 2 integral number power, and if enlargement ratio is 16, then final tache need to be divided by 16 × 16=
256, and 256 can be fast implemented with moving to right for position.Comprehensively consider each side's factor, selects enlargement ratio for 2048.
To acquire the pixel value f (x, y) at target image pixel (x, y), it is assumed that the coordinate is floating number
(26.15,25.66) are first rounded target image coordinate points, are rounded after amplifying to fractional part according to multiplying power, such as following formula:
Wherein, New_x and New_y is the integer part of target point floating-point coordinate value, and Δ x and Δ y are floating number fractional parts
Divide the amplified result rounding value of multiplying power, inv_x and inv_y indicate (1-x) and (1-y), and above data utilizes floor ()
Function is rounded downwards.These parameters are substituted into following formula:
Parameter is integer in formula, and pixel value f of the known source images at (0,0), (0,1), (1,0) and (1,1)
(0,0), f (0,1), f (1,0) and f (1,1) are integer, obtain integer Integer, then displacement fortune to the right is carried out to it
It calculates, it is 22 mobile, achieve the purpose that obtain final interpolation result divided by enlargement ratio 4194304.Shift operation is compared to multiplication and division
Operation has higher efficiency, and shift operation instruction occupies 2 machine cycles, and multiplication and division operational order needs 4 machine cycles,
Requirement while computational efficiency increases to hardware reduces.
By the above method, Mask R-CNN image instance partitioning algorithm is improved, improved algorithm is completed
Under conditions of power equipments defect image and background information are split, the basic of the segmentation of Mask R-CNN had both been remained
Effect, and the speed of bilinear interpolation during RoIAlign is significantly improved, while keeping mapping process sufficiently and uniform
The all pixels of source figure are utilized, segmentation effect is more obvious, and splitting speed is rapider.
From the above technical scheme, this application provides a kind of improvement Mask R-CNN figures for identifying power equipments defect
As example dividing method, comprising: building convolutional neural networks；It reads in power equipments defect and pre-processes picture, power equipment is lacked
It falls into pretreatment picture to be input in convolutional neural networks, convolutional neural networks carry out feature to power equipments defect pretreatment picture
It extracts, obtains the region containing feature；The region containing feature is refined by RPN network, realizes the recurrence to increment,
Region after being refined；In region after refinement, after carrying out RoIAlign operation to refinement by bilinearity difference approach
Region handled, make each fixed-size characteristic pattern of interest Area generation；According to each interest Area generation fixed ruler
Very little characteristic pattern, the classification obtained by Faster-RCNN, coordinate information, and the mask information that full convolutional network obtains, obtain
Divide to power equipments defect example.The application accelerates the ROIAlign of network by improving to bilinear interpolation algorithm
Process, as far as possible reduction interpolation quantity calculation, first by pair of power equipments defect pretreatment picture and target image geometric center
Together, interpolating pixel quantity is needed by the reduction of center alignment thereof, the floating-point operation of Interpolation Process is then converted into shift operation,
Reduce execution cycle.Mask R-CNN image instance dividing method is improved by the above method, improved method exists
It completes under conditions of splitting power equipments defect image and background information, had both remained the segmentation of Mask R-CNN
Basic effect, and the speed of bilinear interpolation during RoIAlign is significantly improved, while keeping mapping process sufficiently and equal
The even all pixels that power equipments defect pretreatment picture is utilized, so that segmentation effect is more obvious, and splitting speed
More rapidly.
The above is only the specific embodiments of the application, it is noted that those skilled in the art are come
It says, under the premise of not departing from the application principle, several improvements and modifications can also be made, these improvements and modifications also should be regarded as
The protection scope of the application.
Claims (4)
1. a kind of improvement Mask R-CNN image instance dividing method for identifying power equipments defect, which is characterized in that method packet
Include following steps:
Step S1: building convolutional neural networks；
Step S2: reading in power equipments defect and pre-process picture, and power equipments defect pretreatment picture is input to convolutional Neural
In network, convolutional neural networks carry out feature extraction to power equipments defect pretreatment picture, obtain the region containing feature；
Step S3: refining the region containing feature by RPN (Region Proposal Network) network, realizes
Recurrence to increment, the region after being refined；
Step S4: in the region after refinement, RoIAlign operation is carried out to the region after refinement by bilinearity difference approach
It is handled, makes each fixed-size characteristic pattern of interest Area generation；
Step S5: according to each fixed-size characteristic pattern of interest Area generation, the classification obtained by Faster-RCNN is sat
Information, and the mask information that full convolutional network obtains are marked, the segmentation of power equipments defect example is obtained.
2. the improvement Mask R-CNN image instance dividing method of identification power equipments defect, feature exist as claimed in claim 1
In, in the region after refinement, by bilinearity difference approach carry out RoIAl ign operation to the region after refinement at
Reason, makes each fixed-size characteristic pattern of interest Area generation include:
Step S41: in the region after refinement, being rounded floating number, will fractional part carry out multiplying power amplification after be rounded into
Row operation, operation result passes through shift right operation again and obtains true fractional part, in addition the operation result of integer part obtains
Interpolation result to the end；
Step S42: RoIAlign operation is carried out by difference result, the region after refinement is handled, make each interest region
Generate fixed-size characteristic pattern.
3. the improvement Mask R-CNN image instance dividing method of identification power equipments defect according to claim 2, feature
It is, lack power equipment when RoIAl ign operation handles the region after refinement by bilinearity difference approach
Fall into the geometric center of pretreatment picture and the geometric center alignment of target image.
4. the improvement Mask R-CNN image instance dividing method of identification power equipments defect, feature exist as claimed in claim 1
In, according to each fixed-size characteristic pattern of interest Area generation, the classification that is obtained by Faster-RCNN, coordinate information, with
And the mask information that full convolutional network obtains, before obtaining the segmentation of power equipments defect example further include:
Classification information and coordinate information are obtained by Faster-RCNN；
Mask information is obtained by full convolutional network.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201910089308.0A CN109816669A (en) | 2019-01-30 | 2019-01-30 | A kind of improvement Mask R-CNN image instance dividing method identifying power equipments defect |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201910089308.0A CN109816669A (en) | 2019-01-30 | 2019-01-30 | A kind of improvement Mask R-CNN image instance dividing method identifying power equipments defect |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| CN109816669A true CN109816669A (en) | 2019-05-28 |
Family
ID=66605750
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201910089308.0A Withdrawn CN109816669A (en) | 2019-01-30 | 2019-01-30 | A kind of improvement Mask R-CNN image instance dividing method identifying power equipments defect |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN109816669A (en) |
Cited By (13)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN110310262A (en) * | 2019-06-19 | 2019-10-08 | 上海理工大学 | A kind of method, apparatus and system for detection wheel tyre defect |
| CN110321933A (en) * | 2019-06-11 | 2019-10-11 | 武汉闻道复兴智能科技有限责任公司 | A kind of fault recognition method and device based on deep learning |
| CN110570410A (en) * | 2019-09-05 | 2019-12-13 | 河北工业大学 | Detection method for automatically identifying and detecting weld defects |
| CN110599497A (en) * | 2019-07-31 | 2019-12-20 | 中国地质大学（武汉） | Drivable region segmentation method based on deep neural network |
| CN110910360A (en) * | 2019-11-14 | 2020-03-24 | 腾讯云计算（北京）有限责任公司 | Power grid image positioning method and image positioning model training method |
| 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 |
| CN111340796A (en) * | 2020-03-10 | 2020-06-26 | 创新奇智(成都)科技有限公司 | Defect detection method and device, electronic equipment and storage medium |
| 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 |
| CN111666811A (en) * | 2020-04-22 | 2020-09-15 | 北京联合大学 | Method and system for extracting traffic sign area in traffic scene image |
| CN112288694A (en) * | 2020-10-19 | 2021-01-29 | 武汉大学 | Mask region convolution neural network-based power transformation equipment defect identification method |
| CN112396620A (en) * | 2020-11-17 | 2021-02-23 | 齐鲁工业大学 | Image semantic segmentation method and system based on multiple thresholds |
| CN112435168A (en) * | 2020-12-01 | 2021-03-02 | 清华大学深圳国际研究生院 | Reference block scaling method and computer-readable storage medium |
| CN113177941A (en) * | 2021-05-31 | 2021-07-27 | 中冶赛迪重庆信息技术有限公司 | Steel coil edge crack identification method, system, medium and terminal |
Citations (5)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN107480730A (en) * | 2017-09-05 | 2017-12-15 | 广州供电局有限公司 | Power equipment identification model construction method and system, the recognition methods of power equipment |
| CN108389207A (en) * | 2018-04-28 | 2018-08-10 | 上海视可电子科技有限公司 | A kind of the tooth disease diagnosing method, diagnostic device and intelligent image harvester |
| CN108921916A (en) * | 2018-07-03 | 2018-11-30 | 广东工业大学 | The painting methods, device in multiple target region, equipment and storage medium in picture |
| CN208172859U (en) * | 2018-04-28 | 2018-11-30 | 上海视可电子科技有限公司 | A kind of intelligent image acquisition device |
| CN109117822A (en) * | 2018-08-31 | 2019-01-01 | 贵州大学 | A kind of part case segmentation recognition method based on deep learning |
-
2019
- 2019-01-30 CN CN201910089308.0A patent/CN109816669A/en not_active Withdrawn
Patent Citations (5)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN107480730A (en) * | 2017-09-05 | 2017-12-15 | 广州供电局有限公司 | Power equipment identification model construction method and system, the recognition methods of power equipment |
| CN108389207A (en) * | 2018-04-28 | 2018-08-10 | 上海视可电子科技有限公司 | A kind of the tooth disease diagnosing method, diagnostic device and intelligent image harvester |
| CN208172859U (en) * | 2018-04-28 | 2018-11-30 | 上海视可电子科技有限公司 | A kind of intelligent image acquisition device |
| CN108921916A (en) * | 2018-07-03 | 2018-11-30 | 广东工业大学 | The painting methods, device in multiple target region, equipment and storage medium in picture |
| CN109117822A (en) * | 2018-08-31 | 2019-01-01 | 贵州大学 | A kind of part case segmentation recognition method based on deep learning |
Non-Patent Citations (1)
| Title |
|---|
| 吴科君: "基于深度学习的海面船舶目标检测", 《中国优秀硕士学位论文全文数据库信息科技辑》 * |
Cited By (19)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN110321933B (en) * | 2019-06-11 | 2021-09-14 | 武汉闻道复兴智能科技有限责任公司 | Fault identification method and device based on deep learning |
| CN110321933A (en) * | 2019-06-11 | 2019-10-11 | 武汉闻道复兴智能科技有限责任公司 | A kind of fault recognition method and device based on deep learning |
| CN110310262A (en) * | 2019-06-19 | 2019-10-08 | 上海理工大学 | A kind of method, apparatus and system for detection wheel tyre defect |
| CN110599497A (en) * | 2019-07-31 | 2019-12-20 | 中国地质大学（武汉） | Drivable region segmentation method based on deep neural network |
| CN110570410A (en) * | 2019-09-05 | 2019-12-13 | 河北工业大学 | Detection method for automatically identifying and detecting weld defects |
| CN110570410B (en) * | 2019-09-05 | 2022-03-22 | 河北工业大学 | Detection method for automatically identifying and detecting weld defects |
| CN110910360A (en) * | 2019-11-14 | 2020-03-24 | 腾讯云计算（北京）有限责任公司 | Power grid image positioning method and image positioning model training method |
| 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 |
| 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 |
| 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 |
| CN111340796A (en) * | 2020-03-10 | 2020-06-26 | 创新奇智(成都)科技有限公司 | Defect detection method and device, electronic equipment and storage medium |
| CN111666811A (en) * | 2020-04-22 | 2020-09-15 | 北京联合大学 | Method and system for extracting traffic sign area in traffic scene image |
| CN111666811B (en) * | 2020-04-22 | 2023-08-15 | 北京联合大学 | Method and system for extracting traffic sign board area in traffic scene image |
| CN112288694A (en) * | 2020-10-19 | 2021-01-29 | 武汉大学 | Mask region convolution neural network-based power transformation equipment defect identification method |
| CN112288694B (en) * | 2020-10-19 | 2022-10-04 | 武汉大学 | Method for identifying defects of power transformation equipment based on mask region convolution neural network |
| CN112396620A (en) * | 2020-11-17 | 2021-02-23 | 齐鲁工业大学 | Image semantic segmentation method and system based on multiple thresholds |
| CN112435168A (en) * | 2020-12-01 | 2021-03-02 | 清华大学深圳国际研究生院 | Reference block scaling method and computer-readable storage medium |
| CN112435168B (en) * | 2020-12-01 | 2024-01-19 | 清华大学深圳国际研究生院 | Reference block scaling method and computer readable storage medium |
| CN113177941A (en) * | 2021-05-31 | 2021-07-27 | 中冶赛迪重庆信息技术有限公司 | Steel coil edge crack identification method, system, medium and terminal |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN109816669A (en) | A kind of improvement Mask R-CNN image instance dividing method identifying power equipments defect | |
| CN107767382B (en) | The extraction method and system of static three-dimensional map contour of building line | |
| JP6705912B2 (en) | Method and apparatus for recognizing character areas in an image | |
| WO2020134082A1 (en) | Path planning method and apparatus, and mobile device | |
| CN103020993B (en) | Visual saliency detection method by fusing dual-channel color contrasts | |
| CN107871321B (en) | Image segmentation method and device | |
| CN111640089A (en) | Defect detection method and device based on feature map center point | |
| EP3460762A1 (en) | Device and method for extracting topographical boundary | |
| WO2023087893A1 (en) | Object processing method and apparatus, computer device, storage medium and program product | |
| CN111291826A (en) | Multi-source remote sensing image pixel-by-pixel classification method based on correlation fusion network | |
| CN103839222A (en) | Grid-to-vector parallel system based on hierarchical boundary topology search model | |
| CN114998097A (en) | Image alignment method, device, computer equipment and storage medium | |
| CN111047682A (en) | Three-dimensional lane model generation method and system | |
| CN111062331A (en) | Mosaic detection method and device for image, electronic equipment and storage medium | |
| CN109685064A (en) | A kind of modified RoIAlign provincial characteristics aggregation algorithms | |
| CN105574903B (en) | A kind of geometric error modeling synthetic method and system accelerated based on GPU | |
| CN109241369A (en) | Rainfall isopleth construction method based on grid stretching method | |
| CN114511862B (en) | Form identification method and device and electronic equipment | |
| Mahmoudpour et al. | Superpixel-based depth map estimation using defocus blur | |
| CN105741335A (en) | Fast drawing method of multi-level war-game map based on blocking structure | |
| CN113487697A (en) | Method and device for generating simple strokes, electronic equipment and storage medium | |
| CN115082597A (en) | Palette-based image recoloring method and system | |
| CN109242750B (en) | Picture signature method, picture matching method, device, equipment and storage medium | |
| CN109215049B (en) | Roof segmentation method, system and equipment based on multi-scale three-dimensional prior information | |
| CN101908062A (en) | Method for judging predicates in GIS (Geographic Information System) space and system thereof |
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 | ||
| WW01 | Invention patent application withdrawn after publication |
Application publication date: 20190528 |
|
| WW01 | Invention patent application withdrawn after publication |