CN111160399A - Deep-learning insulator target sample labeling method and device - Google Patents
Deep-learning insulator target sample labeling method and device Download PDFInfo
- Publication number
- CN111160399A CN111160399A CN201911242358.4A CN201911242358A CN111160399A CN 111160399 A CN111160399 A CN 111160399A CN 201911242358 A CN201911242358 A CN 201911242358A CN 111160399 A CN111160399 A CN 111160399A
- Authority
- CN
- China
- Prior art keywords
- insulator
- marking
- target
- ellipse
- labeling
- 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
- 239000012212 insulator Substances 0.000 title claims abstract description 96
- 238000002372 labelling Methods 0.000 title claims abstract description 39
- 238000013135 deep learning Methods 0.000 title claims abstract description 28
- 238000000034 method Methods 0.000 claims abstract description 21
- 239000011159 matrix material Substances 0.000 claims description 13
- 238000006243 chemical reaction Methods 0.000 claims description 6
- 238000001514 detection method Methods 0.000 abstract description 8
- 238000004880 explosion Methods 0.000 abstract description 5
- 230000002269 spontaneous effect Effects 0.000 abstract description 4
- 238000003062 neural network model Methods 0.000 abstract description 2
- 238000010586 diagram Methods 0.000 description 4
- 230000000694 effects Effects 0.000 description 3
- 238000009413 insulation Methods 0.000 description 2
- 230000004048 modification Effects 0.000 description 2
- 238000012986 modification Methods 0.000 description 2
- 230000002159 abnormal effect Effects 0.000 description 1
- 238000013473 artificial intelligence Methods 0.000 description 1
- 238000005516 engineering process Methods 0.000 description 1
Images
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F18/00—Pattern recognition
- G06F18/40—Software arrangements specially adapted for pattern recognition, e.g. user interfaces or toolboxes therefor
- G06F18/41—Interactive pattern learning with a human teacher
Landscapes
- Engineering & Computer Science (AREA)
- Computer Vision & Pattern Recognition (AREA)
- Theoretical Computer Science (AREA)
- Bioinformatics & Computational Biology (AREA)
- Artificial Intelligence (AREA)
- Bioinformatics & Cheminformatics (AREA)
- Life Sciences & Earth Sciences (AREA)
- Data Mining & Analysis (AREA)
- Evolutionary Biology (AREA)
- Evolutionary Computation (AREA)
- Physics & Mathematics (AREA)
- General Engineering & Computer Science (AREA)
- General Physics & Mathematics (AREA)
- Human Computer Interaction (AREA)
- Insulators (AREA)
Abstract
The invention discloses a method and a device for marking an insulator target sample for deep learning. The invention provides a method for marking insulator sub-targets in an image sample, which is characterized in that an elliptical marking frame is used for marking the positions of insulator targets in the image, and the states of the insulator targets are marked according to categories, such as normal, spontaneous explosion, fouling and the like. Based on the labeling method, the obtained data set can support a deep learning neural network model, end-to-end training is carried out, and a detection model of the insulator and the state (normal/spontaneous explosion/fouling and the like) of the insulator is obtained. The method has the advantages of strong applicability, high labeling efficiency, high labeling precision and the like.
Description
Technical Field
The invention relates to a method and a device for marking an insulator target sample for deep learning.
Background
With the rise of artificial intelligence industry and the increasing maturity of deep learning technology, the end-to-end deep learning target detection model training mode is widely applied.
Even if a person who does not know deep learning can collect the image sample by himself, the image is drawn on the image to label information such as the position, the scale, the type, the state and the like of the target to be detected, and the information of the labeling frame is converted into a group of real numbers to serve as labels of the image sample.
And (3) performing end-to-end deep learning target detection model training by using an end-to-end model, inputting the end-to-end model as an image and outputting the end-to-end model as a label of a sample to obtain a target detection model.
But the mode of sample labeling and the labeling quality both have important influence on the effect of the deep learning target detection model.
In the current common insulator labeling method, in an image, a whole string of insulators is labeled by a rectangular frame, and abnormal insulator parts in the string are specially labeled, as shown in fig. 1.
Some researchers have also marked each insulator individually and also the state using a rectangular frame, as shown in fig. 2. The rectangular frame marking is simple and firm, but a large part of space exists between the boundary of the target to be detected and the boundary of the marking frame, and a large amount of interference information is doped, so that the model effect is poor.
Disclosure of Invention
The invention aims to provide a method for marking an insulator target sample for deep learning, which replaces the existing marking mode.
The invention also aims to provide an insulator target sample marking device for deep learning, so as to replace the existing marking mode.
Therefore, the invention provides a deep learning insulator target sample labeling method, which comprises the following steps: s1, drawing a marking frame in the image sample by using a marking tool, and using a minimum ellipse which can enclose the insulator as the marking frame; s2, converting the ellipse labeling frame data into label data, wherein for each ellipse labeling frame, an abscissa x of the center of the ellipse, a ordinate y of the center, a major semi-axis a, a minor semi-axis b, and an angle t from the x-axis direction to the ellipse major axis direction in a counterclockwise rotation mode are recorded, wherein the range of t is [0, 180); recording the state information of the insulator marked by the oval marking frame by using an integer c; and S3, converting the label information of each insulator target into a structured matrix for each image sample, wherein each row of the structured matrix corresponds to one insulator, and each row records the x, y, a, b, t and c in the step S2.
The invention also provides a device for marking the insulator target sample for deep learning, which comprises the following components: the state marking module is used for drawing a marking frame in the image sample by using a marking tool and using a minimum ellipse which can enclose the insulator as the marking frame; the data conversion module is used for converting the data of the ellipse marking frames into label information, recording an abscissa x of the center of the ellipse, a longitudinal coordinate y of the center, a major semi-axis a, a minor semi-axis b and an angle t from the direction of the x axis to the direction of the major axis of the ellipse by anticlockwise rotation aiming at each ellipse marking frame, wherein the range of t is [0,180 ], and recording the state information of the insulator marked by the ellipse marking frame by using an integer c; and the result storage module is used for converting the label information of each insulator target into a structural matrix for each image sample, wherein each row of the structural matrix corresponds to one insulator, and each row records the x, y, a, b, t and c.
The method for marking the insulator sub-target sample has the following advantages:
1. the model can be trained end to end by utilizing the picture sample and the sample label, and the obtained model can rapidly and intelligently detect the insulator sub-targets and the states thereof in the image;
2. the insulator sub-targets related in the deep learning target detection task are marked by the oval frames, and the method has the advantages of strong applicability, high marking efficiency, high marking precision and the like.
In addition to the objects, features and advantages described above, other objects, features and advantages of the present invention are also provided. The present invention will be described in further detail below with reference to the drawings.
Drawings
The accompanying drawings, which are incorporated in and constitute a part of this application, illustrate embodiments of the invention and, together with the description, serve to explain the invention and not to limit the invention. In the drawings:
FIG. 1 is a schematic diagram of a conventional insulator labeling method;
FIG. 2 is a schematic diagram of a conventional labeling method for a single insulator piece;
FIG. 3 is a schematic diagram of an insulator target labeling method according to the present invention;
FIG. 4 is a flow chart of an insulator target labeling method according to the present invention; and
figure 5 is a block diagram of an insulation sub-target labeling apparatus according to the present invention.
Detailed Description
The present invention will be described in detail below with reference to the embodiments with reference to the attached drawings.
In the process of implementing the invention, the inventor finds that: for an insulator target, its spatial geometry resembles a cylinder, and therefore its boundary tends to be elliptical when projected into a two-dimensional image. Therefore, according to the geometrical characteristics, a more targeted marking method can be used, the adaptability is stronger, and the effect is more excellent.
The invention provides a method for marking insulator sub-targets in an image sample, which is characterized in that an elliptical marking frame is used for marking the positions of insulator targets in the image, and the states of the insulator targets are marked according to categories, such as normal, spontaneous explosion, fouling and the like. Based on the labeling method, the obtained data set can support a deep learning neural network model, end-to-end training is carried out, and a detection model of the insulator and the state (normal/spontaneous explosion/fouling and the like) of the insulator is obtained. The method has the advantages of strong applicability, high labeling efficiency, high labeling precision and the like.
The invention discloses a method for marking an insulator target sample for deep learning, which comprises a method for marking an insulator sub-target in an image sample.
An image sample is an image that may contain insulator sub-targets.
And the sample labeling is to represent the position information, the state information and the like of part or all of the insulator targets in the image sample by a group of real numbers, and the group of real numbers is called a sample label.
The method for marking the insulator sub-targets in the image sample comprises the step of marking the positions of the insulator targets in the image by using an oval marking frame.
The ellipse marking frame is the smallest ellipse frame capable of containing a certain insulating sub-target in the image.
The insulator sub-target refers to the complete insulator that would normally exist there. If the insulator is partially shielded or partially damaged and lost, the shielded or lost part is still a part of the insulator sub-target.
The position marking is to convert the insulator sub-target (or its corresponding marking box) in the image into a set of data indicating the position. For example, the x-axis of the abscissa including the center of the ellipse, the y-axis of the center, the a major axis, the b minor axis, and the t angle rotated counterclockwise from the x-axis direction to the major axis direction of the ellipse, where t ranges from [0,180 ], and 0 ° and 180 ° are the same direction.
The method for marking the insulator sub-targets in the image sample marks the states of the insulator sub-targets by using an integer and establishes a corresponding relation table. For example: 1 indicates normal, 2 indicates burst, 3 indicates fouling, etc.
Examples
The method for marking the sample of the insulator sub-target according to the present invention is described with reference to fig. 3 and 4, taking an image containing a string of insulators with a self-explosion therein as an example.
S1, state labeling
And drawing a marking frame in the image by using a marking tool, marking each insulator by using a minimum ellipse which can enclose the insulator, and marking the state, as shown in fig. 3.
S2, data conversion
And converting the data of the oval labeling box into label data. And each ellipse labeling frame records an abscissa x of the center of the ellipse, an ordinate y of the center, a major semiaxis a, a minor semiaxis b and an angle t from the direction of the x axis to the direction of the major axis of the ellipse, wherein the range of t is [0,180 ]. For the insulator marked by the ellipse marking frame, recording the state information of the insulator by using an integer c, for example: 1 indicates normal, 2 indicates burst, 3 indicates fouling, etc.
S3, storing the result
For the image sample, the label information of the insulator sub-target is a structured matrix. Each row corresponds to one insulator, and each row records x, y, a, b, t and c in S2.
The invention marks the insulation sub-targets related in the deep learning target detection task by adopting the oval frame, and has the advantages of strong applicability, high marking efficiency, high marking precision and the like.
The invention also provides a device for marking the insulator target sample for deep learning, which is shown in fig. 5 and comprises: a state labeling module 10, a data conversion module 20 and a result storage module 30.
The state labeling module 10 is configured to draw a labeling frame in the image sample by using a labeling tool, and use an ellipse that can enclose the insulator as the labeling frame. Preferably, a smallest ellipse that can enclose the insulator piece is used as the marking frame.
The data conversion module 20 is configured to convert the data of the elliptical labeling box into tag information. Preferably, for each ellipse marking frame, recording an abscissa x of the center of the ellipse, an ordinate y of the center, a major semi-axis a, a minor semi-axis b, and an angle t from the x-axis direction to the major axis direction of the ellipse in a counterclockwise rotation manner, wherein the range of t is [0,180 ], and recording the state information of the insulator marked by the ellipse marking frame by using an integer c.
The result storage module 30 is configured to convert, for each image sample, the label information of each insulator target into a structured matrix, where each row of the structured matrix corresponds to one insulator. Preferably, each row of the structured matrix records said x, y, a, b, t, c, respectively.
The above description is only a preferred embodiment of the present invention and is not intended to limit the present invention, and various modifications and changes may be made by those skilled in the art. Any modification, equivalent replacement, or improvement made within the spirit and principle of the present invention should be included in the protection scope of the present invention.
Claims (10)
1. A deep learning insulator target sample labeling method is characterized by comprising the following steps:
s1, drawing a marking frame in the image sample, and using an ellipse enclosed by the insulator as the marking frame;
s2, converting the data in the label box into label information; and
and S3, converting the label information of each insulator target into a structured matrix for each image sample, wherein each row of the structured matrix corresponds to one insulator.
2. The method for labeling insulator target samples for deep learning of claim 1, wherein in step S1, a minimum ellipse enclosing an insulator piece is used as a labeling frame.
3. The method for labeling the insulator target samples for deep learning according to claim 2, wherein in step S2, for each labeling box, the abscissa x of the center of the ellipse, the ordinate y of the center, the major axis a, the minor axis b, and the angle t from the x-axis to the major axis of the ellipse are recorded, wherein t is in the range of [0,180), and the state information of the labeled insulator in the labeling box is recorded by using an integer c.
4. The method for labeling insulator target samples for deep learning of claim 3, wherein in step S3, each row of the structured matrix corresponds to one insulator, and each row records x, y, a, b, t, c.
5. The method for labeling the insulator target sample for deep learning according to claim 1, wherein the image sample is an image containing the insulator sub-target.
6. The method for labeling the insulator target sample for deep learning according to claim 1, wherein the insulator sub-target is a complete insulator which is supposed to exist in the insulator sub-target under normal conditions, and if the insulator is partially blocked or partially damaged and missing, the blocked or missing part is still a part of the insulator sub-target.
7. The utility model provides an insulator target sample mark device of degree of depth study which characterized in that includes:
the state marking module is used for drawing a marking frame in the image sample by using a marking tool and using an ellipse which can enclose the insulator as the marking frame;
the data conversion module is used for converting the data of the oval marking frame into label information;
and the result storage module is used for converting the label information of each insulator target into a structural matrix for each image sample, wherein each row of the structural matrix corresponds to one insulator.
8. The device for marking insulator target samples for deep learning of claim 7, wherein the state marking module uses a smallest ellipse that can enclose an insulator piece as a marking frame.
9. The apparatus for marking insulator target samples for deep learning of claim 8, wherein the data conversion module is configured to record, for each ellipse marking frame, an abscissa x of the center of the ellipse, an ordinate y of the center, a major semi-axis a, a minor semi-axis b, and an angle t of counterclockwise rotation from the x-axis direction to the major axis direction of the ellipse, where t is in the range of [0,180), and record the state information of the insulator marked by the ellipse marking frame by using an integer c.
10. The apparatus for marking insulator target samples for deep learning according to claim 7, wherein the insulator sub-target is a complete insulator which should normally exist therein, and if the insulator is partially shielded or partially damaged and missing, the shielded or missing part is still a part of the insulator sub-target.
Priority Applications (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201911242358.4A CN111160399A (en) | 2019-12-06 | 2019-12-06 | Deep-learning insulator target sample labeling method and device |
Applications Claiming Priority (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201911242358.4A CN111160399A (en) | 2019-12-06 | 2019-12-06 | Deep-learning insulator target sample labeling method and device |
Publications (1)
Publication Number | Publication Date |
---|---|
CN111160399A true CN111160399A (en) | 2020-05-15 |
Family
ID=70556493
Family Applications (1)
Application Number | Title | Priority Date | Filing Date |
---|---|---|---|
CN201911242358.4A Pending CN111160399A (en) | 2019-12-06 | 2019-12-06 | Deep-learning insulator target sample labeling method and device |
Country Status (1)
Country | Link |
---|---|
CN (1) | CN111160399A (en) |
Cited By (1)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN112614094A (en) * | 2020-12-15 | 2021-04-06 | 郑州金惠计算机***工程有限公司 | Insulator string abnormity positioning and identifying method based on sequence state coding |
Citations (2)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN108229473A (en) * | 2017-12-29 | 2018-06-29 | 苏州科达科技股份有限公司 | Vehicle annual inspection label detection method and device |
CN108550177A (en) * | 2018-03-27 | 2018-09-18 | 重庆大学 | A kind of software for marking picture and result being marked to be used for deep learning training |
-
2019
- 2019-12-06 CN CN201911242358.4A patent/CN111160399A/en active Pending
Patent Citations (2)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN108229473A (en) * | 2017-12-29 | 2018-06-29 | 苏州科达科技股份有限公司 | Vehicle annual inspection label detection method and device |
CN108550177A (en) * | 2018-03-27 | 2018-09-18 | 重庆大学 | A kind of software for marking picture and result being marked to be used for deep learning training |
Non-Patent Citations (4)
Title |
---|
付征: "基于视频的人脸识别***研究与实现", vol. 2019, no. 11, pages 27 - 28 * |
胡建平 等: "一种新的航拍玻璃绝缘子图像分割方法", 东北电力大学学报, vol. 38, no. 2, pages 91 * |
虢韬 等: "基于Faster RCNN的绝缘子自爆缺陷识别", 电瓷避雷器, pages 184 * |
陈芳 等: "绝缘子串红外图像中单个绝缘子盘面的提取方法", 电网技术, vol. 34, no. 5, pages 221 - 222 * |
Cited By (1)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN112614094A (en) * | 2020-12-15 | 2021-04-06 | 郑州金惠计算机***工程有限公司 | Insulator string abnormity positioning and identifying method based on sequence state coding |
Similar Documents
Publication | Publication Date | Title |
---|---|---|
CN110033453B (en) | Power transmission and transformation line insulator aerial image fault detection method based on improved YOLOv3 | |
CN107358194B (en) | Violence sorting express delivery judgment method based on computer vision | |
CN101589408B (en) | Marker generating and marker detecting system, method and program | |
CN110570433B (en) | Image semantic segmentation model construction method and device based on generation countermeasure network | |
CN114445706A (en) | Power transmission line target detection and identification method based on feature fusion | |
CN107506406B (en) | Picture archiving method and system for power transmission line tower | |
CN107480203A (en) | It is a kind of to be directed to identical and similar pictures duplicate removal view data cleaning method | |
CN115131444A (en) | Calibration method based on monocular vision dispensing platform | |
CN109048903B (en) | Method and equipment for inserting mechanical arm into shaft-inserting hole | |
CN109146880A (en) | A kind of electric device maintenance method based on deep learning | |
CN111160399A (en) | Deep-learning insulator target sample labeling method and device | |
CN104992173A (en) | Symbol recognition method and system used for medical report | |
CN106650831A (en) | Damage detecting method and apparatus based on feature fusion | |
CN110310278A (en) | Sliding window scan method based on surplus patterning method | |
CN108873363B (en) | Three-dimensional light field imaging system and method based on structured light marking | |
CN113189630B (en) | Large-scale construction site personnel identification and positioning method for tower crane installation monitoring | |
CN109815199A (en) | A kind of choosing method and device of picture compression algorithm | |
CN110674827A (en) | Equipment state visual detection method integrating deep learning and morphological filtering | |
CN116089944A (en) | Cross-platform application program abnormality detection method and system based on transfer learning | |
CN205015734U (en) | Stage equipment and performance scheme calibrating device | |
CN110534071A (en) | A kind of display color calibration system neural network based and method | |
CN112347889B (en) | Substation operation behavior identification method and device | |
CN114549282A (en) | Method and system for realizing multi-meter reading based on affine transformation | |
CN109241231A (en) | The accurately pretreatment unit and method of diagram data | |
CN110827259B (en) | Sample augmentation method and device for target key point detection |
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 |