CN109615611B - Inspection image-based insulator self-explosion defect detection method - Google Patents
Inspection image-based insulator self-explosion defect detection method Download PDFInfo
- Publication number
- CN109615611B CN109615611B CN201811376353.6A CN201811376353A CN109615611B CN 109615611 B CN109615611 B CN 109615611B CN 201811376353 A CN201811376353 A CN 201811376353A CN 109615611 B CN109615611 B CN 109615611B
- Authority
- CN
- China
- Prior art keywords
- insulator
- self
- explosion
- inspection image
- defect
- 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.)
- Active
Links
- 239000012212 insulator Substances 0.000 title claims abstract description 305
- 238000004880 explosion Methods 0.000 title claims abstract description 227
- 230000007547 defect Effects 0.000 title claims abstract description 226
- 238000007689 inspection Methods 0.000 title claims abstract description 143
- 238000001514 detection method Methods 0.000 title claims abstract description 89
- 239000011521 glass Substances 0.000 claims abstract description 62
- 238000012549 training Methods 0.000 claims abstract description 50
- 238000001914 filtration Methods 0.000 claims abstract description 48
- 238000000034 method Methods 0.000 claims abstract description 40
- 238000013135 deep learning Methods 0.000 claims abstract description 37
- 238000012545 processing Methods 0.000 claims abstract description 17
- 238000002372 labelling Methods 0.000 claims description 11
- 230000002950 deficient Effects 0.000 claims description 10
- 238000009877 rendering Methods 0.000 claims description 8
- 230000000007 visual effect Effects 0.000 claims description 3
- 238000004088 simulation Methods 0.000 abstract description 2
- 238000013527 convolutional neural network Methods 0.000 description 25
- 238000000605 extraction Methods 0.000 description 9
- 239000013598 vector Substances 0.000 description 9
- 230000006870 function Effects 0.000 description 7
- 238000004364 calculation method Methods 0.000 description 6
- 238000012360 testing method Methods 0.000 description 4
- 238000012952 Resampling Methods 0.000 description 3
- 238000012937 correction Methods 0.000 description 3
- 230000000694 effects Effects 0.000 description 3
- 238000005516 engineering process Methods 0.000 description 3
- 238000012423 maintenance Methods 0.000 description 3
- 230000005540 biological transmission Effects 0.000 description 2
- 238000004422 calculation algorithm Methods 0.000 description 2
- 238000003384 imaging method Methods 0.000 description 2
- 230000001965 increasing effect Effects 0.000 description 2
- 238000013528 artificial neural network Methods 0.000 description 1
- 125000004122 cyclic group Chemical group 0.000 description 1
- 238000003708 edge detection Methods 0.000 description 1
- 230000002708 enhancing effect Effects 0.000 description 1
- 238000009434 installation Methods 0.000 description 1
- 238000004519 manufacturing process Methods 0.000 description 1
- 238000013507 mapping Methods 0.000 description 1
- 238000005065 mining Methods 0.000 description 1
- 238000007781 pre-processing Methods 0.000 description 1
Images
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T7/00—Image analysis
- G06T7/0002—Inspection of images, e.g. flaw detection
- G06T7/0004—Industrial image inspection
- G06T7/0006—Industrial image inspection using a design-rule based approach
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T5/00—Image enhancement or restoration
- G06T5/70—Denoising; Smoothing
-
- 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
- G06T2207/20024—Filtering details
- G06T2207/20032—Median filtering
-
- 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
- G06T2207/30—Subject of image; Context of image processing
- G06T2207/30108—Industrial image inspection
- G06T2207/30164—Workpiece; Machine component
-
- Y—GENERAL TAGGING OF NEW TECHNOLOGICAL DEVELOPMENTS; GENERAL TAGGING OF CROSS-SECTIONAL TECHNOLOGIES SPANNING OVER SEVERAL SECTIONS OF THE IPC; TECHNICAL SUBJECTS COVERED BY FORMER USPC CROSS-REFERENCE ART COLLECTIONS [XRACs] AND DIGESTS
- Y04—INFORMATION OR COMMUNICATION TECHNOLOGIES HAVING AN IMPACT ON OTHER TECHNOLOGY AREAS
- Y04S—SYSTEMS INTEGRATING TECHNOLOGIES RELATED TO POWER NETWORK OPERATION, COMMUNICATION OR INFORMATION TECHNOLOGIES FOR IMPROVING THE ELECTRICAL POWER GENERATION, TRANSMISSION, DISTRIBUTION, MANAGEMENT OR USAGE, i.e. SMART GRIDS
- Y04S10/00—Systems supporting electrical power generation, transmission or distribution
- Y04S10/50—Systems or methods supporting the power network operation or management, involving a certain degree of interaction with the load-side end user applications
Landscapes
- Engineering & Computer Science (AREA)
- Physics & Mathematics (AREA)
- General Physics & Mathematics (AREA)
- Theoretical Computer Science (AREA)
- Quality & Reliability (AREA)
- Computer Vision & Pattern Recognition (AREA)
- Investigating Materials By The Use Of Optical Means Adapted For Particular Applications (AREA)
- Image Analysis (AREA)
- Image Processing (AREA)
Abstract
The invention provides an insulator self-explosion defect detection method based on inspection images, which comprises the following steps: acquiring an inspection image to be detected, which is shot by an electric line inspection unmanned aerial vehicle and comprises a glass insulator, and manually marking a small amount of insulator self-explosion defects; generating virtual inspection images containing insulator defects in batches in a simulation system; performing filtering enhancement processing on the real and virtual images to obtain a patrol image combined training set after filtering enhancement; acquiring an insulator self-explosion defect detection model based on a fast R-CNN deep learning network by utilizing combined training set training; the detection model can detect the filtered and enhanced inspection image to be detected, and whether the inspection image to be detected comprising the glass insulator comprises a self-explosion defect or not is determined. The method solves the problem of insufficient quantity of big data training sets in the prior deep learning training; the accuracy of insulator self-explosion defect detection is improved.
Description
Technical Field
The invention relates to the technical field of operation and maintenance of power grids, in particular to an insulator self-explosion defect detection method based on inspection images.
Background
The glass insulator is widely applied in the operation of a power grid, and is a key shooting part for inspection of an unmanned aerial vehicle. Concentrated self-explosion of the glass insulator may cause line tripping faults, endangering stable operation of the line.
In recent years, the electric line inspection unmanned aerial vehicle has been popularized and applied, and mass inspection images are generated. The inspection image is read manually and the insulator self-explosion defect is detected, so that the efficiency is low, the subjectivity is strong, the consistency is poor, and the operation and maintenance work efficiency is greatly limited. The existing method for automatically detecting the insulator self-explosion defect based on the inspection image cannot meet the application requirements in terms of accuracy and omission factor.
Disclosure of Invention
The invention provides an insulator self-explosion defect detection method based on a patrol image, which aims to solve the problem that the accuracy and the omission factor of the existing insulator self-explosion defect detection method cannot meet the application requirements.
The invention provides an insulator self-explosion defect detection method for inspection images, which comprises the following steps:
acquiring a plurality of real inspection images shot by an electric line inspection unmanned aerial vehicle and comprising defects of a glass insulator, and manually marking the defects of the insulator self-explosion;
generating a virtual inspection image containing the insulator self-explosion defect and automatically marking the insulator self-explosion defect;
Performing filtering enhancement processing on the real patrol image and the virtual patrol image to obtain a real patrol image and a virtual patrol image after filtering enhancement;
combining the real inspection image and the virtual inspection image after the filtering enhancement according to a preset quantity ratio to form a combined training set;
and utilizing a combined training set to train based on a fast R-CNN deep learning network to determine an insulator self-explosion defect detection model, wherein when the insulator self-explosion defect detection model determines the self-explosion defect type of the insulator in a real inspection image or a virtual inspection image, the confidence coefficient of the corresponding self-explosion defect type is larger than a preset detection threshold value.
Further, the self-explosion defect types include: single-chip self-explosion in the glass insulator string, continuous multi-chip self-explosion in the glass insulator string, single-chip self-explosion at the hardware fitting connecting end in the glass insulator string and multi-chip self-explosion at the hardware fitting connecting end of the glass insulator string.
Further, the insulator self-explosion defect detection method further comprises the following steps:
acquiring an inspection image to be detected, which is shot by an electric line inspection unmanned aerial vehicle and comprises a glass insulator;
performing filtering enhancement treatment on the inspection image to be detected to obtain the inspection image to be detected after filtering enhancement;
And detecting by using an insulator self-explosion defect detection model, and filtering the reinforced inspection image to be detected to determine whether the inspection image to be detected comprising the glass insulator comprises the self-explosion defect.
Further, performing filter enhancement processing, including:
1) Traversing the inspection image by adopting four-neighborhood intermediate value filtering to obtain an inspection image after median filtering;
2) Traversing the median filtered inspection image by adopting a Sobel edge operator with 4 direction templates of 0 degree, 45 degrees, 90 degrees and 135 degrees to obtain an inspection image with a protruding edge;
3) And denoising the inspection image to be detected with the edge protruding by using a four-neighborhood gray scale contrast method to obtain the inspection image after the filtering enhancement.
Further, the denoising method for the inspection image to be detected after the edge is protruded by using the four-neighborhood gray scale contrast method, so as to obtain the inspection image to be detected after the filtering enhancement, comprising the following steps:
for any pixel point f (x, y), if |f (x, y) -f (x+i, y+j) | > TH, for any i= -1,1; j= -1,1 are all true, then the pixel point is the noise point; wherein the threshold TH is a constant of not less than 20;
the gray value of the pixel point determined as the noise point is designated as 255.
Further, generating a virtual inspection image containing the insulator self-explosion defect and automatically labeling the insulator self-explosion defect comprises the following steps:
1) Measuring real objects of the glass insulator with different self-explosion defect types and the normal glass insulator by using a laser scanner to obtain three-dimensional geometric data of the real objects;
2) According to the three-dimensional geometric data, respectively establishing a three-dimensional virtual model aiming at the insulator sheets, the normal insulator sheets and the connecting pieces of different self-explosion defect types, and rendering to obtain defective insulator images, normal insulator images and images of the hardware connecting pieces;
3) Generating an insulator string virtual image according to a preset rule, wherein at least one defective insulator is randomly distributed in different areas of the insulator string virtual image;
4) And simulating and photographing the generated virtual images of the insulator strings from a plurality of different angles, rendering and outputting photographing results into a picture image format, and recording coordinate information of pixel points of the areas where the defective insulators and the adjacent normal insulators are located.
Further, manually marking the insulator self-explosion defect includes:
and manually marking the self-explosion defect areas in the obtained real inspection images of the glass insulators with different self-explosion defect types and the areas where the adjacent normal insulator sheets or the connecting ends of the hardware fittings are located.
Further, the method comprises the steps of,
The automatic labeling of the insulator self-explosion defect comprises the following steps:
and determining coordinate information of pixel points of areas where the self-explosion defect insulator sheets and adjacent normal insulator sheets or hardware fitting connecting ends of the self-explosion defect insulator sheets are located in virtual marks of the glass insulators with different self-explosion defect types under a plurality of different shooting visual angles.
Further, the method further comprises the following steps:
after the fact that the to-be-detected inspection image comprising the glass insulator comprises the self-explosion defect is determined, the minimum circumscribed rectangle, attached to the insulator sheet with the self-explosion defect and the pixel area where the normal insulators adjacent to the two ends of the insulator sheet are located, is overlapped on the to-be-detected inspection image.
Further, when the combined training set is used for training based on the fast R-CNN deep learning network to determine the insulator self-explosion defect detection model,
if the confidence coefficient aiming at the corresponding self-explosion defect type is larger than a preset detection threshold value, determining that the detection result is effective, and marking the area where the insulator self-explosion defect is located by using a rectangular frame in the inspection image;
if the confidence coefficient aiming at the corresponding self-explosion defect type is not greater than a preset detection threshold value, determining that the detection result is invalid.
According to the method for detecting the insulator self-explosion defect of the inspection image, provided by the invention, the filter is utilized to enhance and highlight the target geometric characteristics and the fitting of the characteristics in the virtual image and the real image. The combined training set is adopted to train the fast R-CNN model, so that the problem of insufficient quantity of big data training sets in the previous deep learning training is solved, and meanwhile, the labor cost for manually marking the defect targets in the inspection images on a large scale is reduced.
Drawings
Exemplary embodiments of the present invention may be more completely understood in consideration of the following drawings:
fig. 1 is a schematic flow chart of an inspection image-based method for detecting an insulator self-explosion defect according to a preferred embodiment of the present invention;
FIG. 2 is a schematic flow chart of a filtering enhancement step in an inspection image-based method for detecting an insulator self-explosion defect according to a preferred embodiment of the present invention;
fig. 3 is a flow chart of an insulator self-explosion defect detection method based on inspection images in a preferred embodiment of the invention.
Detailed Description
The exemplary embodiments of the present invention will now be described with reference to the accompanying drawings, however, the present invention may be embodied in many different forms and is not limited to the examples described herein, which are provided to fully and completely disclose the present invention and fully convey the scope of the invention to those skilled in the art. The terminology used in the exemplary embodiments illustrated in the accompanying drawings is not intended to be limiting of the invention. In the drawings, like elements/components are referred to by like reference numerals.
Unless otherwise indicated, terms (including technical and scientific terms) used herein have the same meaning as commonly understood by one of ordinary skill in the art. In addition, it will be understood that terms defined in commonly used dictionaries should be interpreted as having a meaning that is consistent with their meaning in the context of the relevant art and will not be interpreted in an idealized or overly formal sense.
When the defect detection of the power equipment is carried out based on the inspection images, two problems cannot be solved completely, namely the defect problem of the inspection equipment defect image data set acquired by each platform is solved, and the self-explosion defect position of the insulator is accurately positioned from the similar insulator components in the normal state. The detection of the insulator self-explosion defect position in the inspection image obtained by the electric unmanned aerial vehicle has the two problems.
Therefore, the deep learning technology is adopted to develop the self-explosion detection of the glass insulator based on the inspection image, which has important significance. The deep learning technology inevitably involves training the deep learning network. When training the deep learning network, a large number of marked insulator self-explosion defect image standard samples are required to be used as a training set and a testing set of the deep learning network.
In order to obtain a sufficient number of marked insulator self-explosion defect image standard samples, an electric unmanned aerial vehicle is required to obtain the self-explosion defect standard images of various types of insulators under different shooting angles, and the self-explosion defect standard images are manually marked with information. Both the time and labor costs of these two tasks are very high.
Another concern is the fact that, although the resolution of the inspection image of the electrical equipment acquired by the current electrical unmanned aerial vehicle is higher and higher, the range of the area of the self-explosion defect image of the insulator in the whole inspection image is limited by the shooting distance, and the characteristic information of the self-explosion defect image is easily submerged in the surrounding environment and is easily interfered by the nearby normal insulator, so that even if the manual operation is performed, it is difficult to accurately locate the insulator target of the self-explosion defect from the inspection image.
In addition, because the occurrence rate of the self-explosion defects of the glass insulators in the power system is low, the number of the glass insulators based on the self-explosion defects is small, and it is difficult to finish multi-angle large-batch acquisition of inspection images containing the self-explosion defects of the insulators, and the number of defect samples of target equipment is difficult to meet the data requirements of deep learning network training and testing. The insufficient number of the defect samples leads to the occurrence of an overfitting phenomenon of the generated detection model, so that the identification effect is unstable.
Therefore, for the deep learning technology with the best performance in the field of target detection, the characteristic information of the insulator self-explosion target with smaller size can be lost after the convolution processing and the downsampling layer of the patrol image with original resolution due to the limitation of the capacity of the existing neural network, and the insulator self-explosion target is more difficult to identify.
The invention provides an insulator self-explosion defect detection method based on inspection images, which comprises the following steps: acquiring an inspection image to be detected, which is shot by an electric line inspection unmanned aerial vehicle and comprises a glass insulator, and manually marking a small amount of insulator self-explosion defects; generating virtual inspection images containing insulator defects in batches in a simulation system; performing filtering enhancement processing on the real and virtual images to obtain a patrol image combined training set after filtering enhancement; acquiring an insulator self-explosion defect detection model based on a fast R-CNN deep learning network by utilizing combined training set training; the detection model can detect the filtered and enhanced inspection image to be detected, and whether the inspection image to be detected comprising the glass insulator comprises a self-explosion defect or not is determined.
The invention provides a complete technical route of generating a virtual sample, automatically marking a target, forming a combined training set, filtering and enhancing pretreatment, training a deep learning detection model. The method provided by the invention solves the problem of insufficient quantity of big data training sets in the prior deep learning training by utilizing virtual inspection images generated in batches, and simultaneously reduces the labor cost for manually marking the defect targets in the inspection images on a large scale; the method comprises the steps of adopting filtering to enhance and highlight the fitting of target geometric features and features in virtual images and real images; and the combined training set is adopted to train the fast R-CNN model, so that the accuracy of insulator self-explosion defect detection is improved.
As shown in fig. 1, the method for detecting the self-explosion defect of the insulator of the inspection image according to the embodiment of the invention comprises the following steps:
s10: acquiring a plurality of real inspection images shot by an electric line inspection unmanned aerial vehicle and comprising defects of a glass insulator, and manually marking the defects of the insulator self-explosion;
s20: generating a virtual inspection image containing the insulator self-explosion defect and automatically marking the insulator self-explosion defect;
s30: performing filtering enhancement processing on the real patrol image and the virtual patrol image to obtain a real patrol image and a virtual patrol image after filtering enhancement;
s40: combining the real inspection image and the virtual inspection image after the filtering enhancement according to a preset quantity ratio to form a combined training set;
s50: and utilizing a combined training set to train based on a fast R-CNN deep learning network to determine an insulator self-explosion defect detection model, wherein when the insulator self-explosion defect detection model determines the self-explosion defect type of the insulator in a real inspection image or a virtual inspection image, the confidence coefficient of the corresponding self-explosion defect type is larger than a preset detection threshold value.
It should be understood that the plurality of real inspection images taken by the power line inspection unmanned aerial vehicle, including defects of the glass insulator, includes various typical defect types. It should be understood that, when determining the type of the self-explosion defect of the insulator in the real inspection image or the virtual inspection image, the confidence of the self-explosion defect detection model for the corresponding type of the self-explosion defect is greater than a preset detection threshold; if the confidence coefficient is not greater than the preset detection threshold value, further adjusting the structural parameter or the weight parameter of the self-explosion defect type until the confidence coefficient for detecting all the inspection images with typical defect types is greater than the preset detection threshold value.
In specific implementation, the method for detecting the self-explosion defect of the insulator further comprises the following steps:
acquiring an inspection image to be detected, which is shot by an electric line inspection unmanned aerial vehicle and comprises a glass insulator;
performing filtering enhancement treatment on the inspection image to be detected to obtain the inspection image to be detected after filtering enhancement;
and detecting by using an insulator self-explosion defect detection model, and filtering the reinforced inspection image to be detected to determine whether the inspection image to be detected comprising the glass insulator comprises the self-explosion defect.
It should be understood that the inspection image to be detected including the glass insulator may or may not include a self-explosion defect.
The types of the self-explosion defects possibly included in the inspection image to be detected including the glass insulator include: single-chip self-explosion in the glass insulator string, continuous multi-chip self-explosion in the glass insulator string, single-chip self-explosion at the hardware fitting connecting end in the glass insulator string and multi-chip self-explosion at the hardware fitting connecting end of the glass insulator string.
In specific implementation, the method for detecting the self-explosion defect of the insulator carries out filtering enhancement treatment and comprises the following steps:
1) Traversing the inspection image by adopting four-neighborhood intermediate value filtering to obtain an inspection image after median filtering;
2) Traversing the median filtered inspection image by adopting a Sobel edge operator with 4 direction templates of 0 degree, 45 degrees, 90 degrees and 135 degrees to obtain an inspection image with a protruding edge;
3) And denoising the inspection image to be detected with the edge protruding by using a four-neighborhood gray scale contrast method to obtain the inspection image after the filtering enhancement.
In specific implementation, the insulator self-explosion defect detection method uses a four-neighborhood gray scale contrast method to denoise the inspection image to be detected after edge protrusion, and obtains the inspection image to be detected after filtering enhancement, and the method comprises the following steps:
for any pixel point f (x, y), if |f (x, y) -f (x+i, y+j) | > TH, for any i= -1,1; j= -1,1 are all true, then the pixel point is the noise point; wherein the threshold TH is a constant of not less than 20;
the gray value of the pixel point determined as the noise point is designated as 255.
In specific implementation, the method for detecting the self-explosion defect of the insulator generates a virtual inspection image containing the self-explosion defect of the insulator and automatically marks the self-explosion defect of the insulator, and comprises the following steps:
1) Measuring real objects of the glass insulator with different self-explosion defect types and the normal glass insulator by using a laser scanner to obtain three-dimensional geometric data of the real objects;
2) According to the three-dimensional geometric data, respectively establishing a three-dimensional virtual model aiming at the insulator sheets, the normal insulator sheets and the connecting pieces of different self-explosion defect types, and rendering to obtain defective insulator images, normal insulator images and images of the hardware connecting pieces;
3) Generating an insulator string virtual image according to a preset rule, wherein at least one defective insulator is randomly distributed in different areas of the insulator string virtual image;
4) And simulating and photographing the generated virtual images of the insulator strings from a plurality of different angles, rendering and outputting photographing results into a picture image format, and recording coordinate information of pixel points of the areas where the defective insulators and the adjacent normal insulators are located.
In specific implementation, in the method for detecting the self-explosion defect of the insulator, manually marking the self-explosion defect of the insulator comprises the following steps:
and manually marking the self-explosion defect areas in the obtained real inspection images of the glass insulators with different self-explosion defect types and the areas where the adjacent normal insulator sheets or the connecting ends of the hardware fittings are located.
In specific implementation, in the method for detecting the self-explosion defect of the insulator, the automatic labeling of the self-explosion defect of the insulator comprises the following steps:
and determining coordinate information of pixel points of areas where the self-explosion defect insulator sheets and adjacent normal insulator sheets or hardware fitting connecting ends of the self-explosion defect insulator sheets are located in virtual marks of the glass insulators with different self-explosion defect types under a plurality of different shooting visual angles.
After marking, the defect area is enlarged, and a defect combination area is formed, so that the target detection is facilitated. Specifically, the defect combining area includes the following types:
A single insulator sheet with self-explosion defects and a set of pixel points where normal insulator sheets adjacent to two ends of the single insulator sheet are located; or (b)
A plurality of insulator sheets with continuous self-explosion defects, and a set of pixel points where two normal insulator sheets adjacent to the plurality of insulator sheets at two ends respectively are located;
the hardware fitting connecting end, a first insulator with self-explosion defects, which is connected with the hardware fitting, and a set of pixel points where the next normal insulator piece adjacent to the first insulator piece is located;
the device comprises a hardware fitting connecting end, a plurality of continuous insulators with self-explosion defects, which are connected with the hardware fitting, and a collection of pixel points where the next normal insulator sheet adjacent to the plurality of insulators is located.
In specific implementation, the method for detecting the self-explosion defect of the insulator further comprises the following steps:
after the fact that the to-be-detected inspection image comprising the glass insulator comprises the self-explosion defect is determined, the minimum circumscribed rectangle, attached to the insulator sheet with the self-explosion defect and the pixel area where the normal insulators adjacent to the two ends of the insulator sheet are located, is overlapped on the to-be-detected inspection image.
In the specific implementation, the insulator self-explosion defect detection method,
when the deep learning network based on the Faster R-CNN is trained by utilizing the combined training set and the insulator self-explosion defect detection model is determined,
If the confidence coefficient aiming at the corresponding self-explosion defect type is larger than a preset detection threshold value, determining that the detection result is effective, and marking the area where the insulator self-explosion defect is located by using a rectangular frame in the inspection image;
if the confidence coefficient aiming at the corresponding self-explosion defect type is not greater than a preset detection threshold value, determining that the detection result is invalid.
The method for detecting the self-explosion defects of the insulators, which is provided by the embodiment, takes the self-explosion defects of the glass insulators in the inspection images of the electric unmanned aerial vehicle as detection targets, and utilizes the characteristic that the insulators with the self-explosion defects and the adjacent normal insulators have rich structural characteristics to establish a combined identification model and serve as targets to be detected of the self-explosion defects of the insulators; adopting a virtual engine to render data and a small amount of real tour inspection images to jointly construct a training set; the original image data is used for training a fast R-CNN deep learning network after being filtered and enhanced, and a detection model with accuracy meeting the requirement is generated; the generated detection model is used for processing the inspection image acquired by the electric unmanned aerial vehicle, so that the insulator self-explosion defect can be effectively detected.
The insulator self-explosion defect detection method provided by the embodiment improves the accuracy of insulator self-explosion detection, so that the operation and maintenance working efficiency of the power transmission line can be effectively improved.
The method for detecting the self-explosion defect of the insulator provided by the embodiment is described in detail below.
1) Manufacturing a virtual insulator three-dimensional model
1.1 insulator target region selection
Because the relative range of the detection target of the insulator sheet with the self-explosion defect in the power inspection image is smaller, the embodiment expands the detection target to the insulator sheet with the self-explosion defect and the normal insulator sheets with the adjacent two ends.
The combined area occupied by the insulator sheet with the self-explosion defect and the normal insulator sheets adjacent to the two ends of the insulator sheet has proper size in the inspection image, has clear characteristics and is relatively stable, and the combined area is used as the input of a detection model generated based on a deep learning network, so that the detection accuracy can be improved.
Specifically, there are the following 3 cases in which the combined area occupied by the insulator sheet having the self-explosion defect and the normal insulator sheet adjacent to both ends thereof is:
in the insulator string, the area of the insulator sheet with the self-explosion defect is the position of the self-explosion defect; if two adjacent insulator sheets on the left side and the right side are normal, the normal two insulator sheets are taken as the boundary of the combined area with the self-explosion defect.
If a plurality of continuous insulator sheets with self-explosion defects exist in the insulator string, the two normal insulator sheets are taken as the boundary of the combined area with the self-explosion defects until two adjacent insulators on the left side and the right side of the insulator string are normal.
If the first insulator at the joint of the insulator string and the hardware fitting has the self-explosion defect, the boundary of the combined area with the self-explosion defect is formed from the connecting end of the self-explosion insulator and the hardware fitting to the adjacent normal insulator sheet.
In generating the detection model based on the deep learning network, each standard image sample in the training set or the detection used comprises at least one of the above 3 types of combined regions.
1.2 definition classification of insulator self-explosion defect targets
The defect sample in this embodiment includes four types of glass insulator self-explosion defect targets, namely single-chip self-explosion in the glass insulator string, continuous multi-chip self-explosion in the glass insulator string, single-chip self-explosion at the hardware fitting connecting end in the glass insulator string and multi-chip self-explosion at the hardware fitting connecting end of the glass insulator string, and the glass insulator string which is normally installed is used as a comparison sample for performing a subsequent defect detection test.
1.3 three-dimensional laser scanning of real insulator components
The insulator and the connecting piece thereof are accurately measured by a laser scanner, so that accurate modeling data of the insulator are obtained.
Specifically, a laser scanner is used for respectively scanning standard samples of four types of glass insulator self-explosion defects such as single-chip self-explosion in the glass insulator string, continuous multi-chip self-explosion in the glass insulator string, single-chip self-explosion at the connecting end of the glass insulator string, multi-chip self-explosion at the connecting end of the glass insulator string and the like and the glass insulator string which is used as a reference and is normally installed, so that accurate three-dimensional geometric information of various insulators which are normal and defective is obtained.
1.4 virtual sample modeling
Specifically, 3DMax software with stronger modeling function and rendering capability is adopted when the virtual sample is modeled. Inputting accurate three-dimensional information of each part of the insulator obtained by using a laser scanner, and modeling the insulator sheet and the corresponding connecting hardware fitting by using a polygonal method to obtain an accurate three-dimensional insulator combination and an insulator self-exposure defect model.
Geomagic software can also be used for data processing and three-dimensional modeling. And a Unity3D rendering component is adopted to enable the model to achieve the effect close to reality.
In addition, for insulator self-explosion defects, functions need to be designed so that four types of defects are randomly generated in different areas of the virtual insulator string.
It should be appreciated that an effective virtual image sample includes at least one of the 4 defect types.
1.5 generating a virtual inspection image sample containing the insulator self-explosion defect
In unmanned aerial vehicle power inspection, unmanned aerial vehicle usually flies along overhead transmission lines, and images of inspection of target equipment are taken at prescribed working positions. When the virtual inspection image sample is generated, the virtual camera in the virtual scene is controlled according to the shooting characteristics of the unmanned aerial vehicle inspection, so that the shooting angle in the real unmanned aerial vehicle power inspection is simulated, and the generated virtual sample is maximally close to the real unmanned aerial vehicle inspection image.
Specifically, by means of a Camera (Camera) function in the Unity3D game engine, analog photographing is carried out on the insulator string, and a photographing result is rendered and output into a universal picture image format; and recording pixel coordinate information of the insulator defect position in the virtual image and generating a corresponding labeling information file of the insulator self-explosion part, wherein the labeling information file is recorded with defect attributes of the insulator self-explosion defect and pixel coordinate values at four vertexes of a rectangular area where the self-explosion position is located.
The method for generating the virtual image and the labeling information solves the problem that the insulator self-explosion defect samples in the inspection image are rare, and greatly reduces the workload of manual labeling.
Combining the virtual image insulator self-explosion defect sample and the real image insulator self-explosion defect sample acquired by the electric unmanned aerial vehicle according to a set proportion, and jointly using the virtual image insulator self-explosion defect sample and the real image insulator self-explosion defect sample as a training set or a test set when a detection model is generated based on a deep learning network.
2) Image filtering enhancement
The insulator self-explosion defect targets have relatively outstanding geometric outline characteristics, and the embodiment introduces a filtering enhancement step to further enhance the edge gradient information of the insulator self-explosion defect targets in the inspection image, and meanwhile weakens the influence of background noise on the detection result so as to increase the detection probability of the insulator self-explosion defect targets. The specific steps of image filtering enhancement are shown in fig. 2.
2.1. Median filtering of images
Both edges and noise in the image appear as high frequency components in the frequency domain. The image is filtered prior to detection to reduce the effect of noise on edge detection.
Preferably, median filtering is employed. When in median filtering, a window with a fixed size is used for scanning point by point on an image, pixels in the window are arranged according to the gray scale, and a gray scale value in the middle of the gray scale is taken as the gray scale value of the pixels in the center of the window.
Specifically, for an inspection image with a resolution of mxn, the gray value of each pixel in the inspection image is recorded as f (x, y), the image is traversed by using a four-neighborhood median filter, and after median filtering, the gray value of each pixel in the inspection image is recorded as G (x, y):
G(x,y)＝MED{f(x-1,y),f(x,y-1),f(x,y+1),f(x+1,y),f(x,y)}
wherein MED is a function of the median.
2.2. Increasing Sobel direction templates
There are many directions of the edges of the image, and other edge directions, such as 45, in addition to the horizontal and vertical directions.
In order to increase the accuracy of detecting the edge of the pixel point, the direction templates of the classical Sobel edge operator are increased from 2 to 4, and the directions are respectively:
and traversing each inspection image by using the 4 direction templates respectively. For example, in the resampling operation, the resampling value of the gray scale is the gray scale value calculated by weighting the pixel values in the 3×3 region with the resampling value as the center point according to the weight of the template.
2.3 noise Point cancellation
All potential targets in the insulator self-explosion defect image have two characteristics of direction and amplitude at the edge. The gray value of the pixel is changed severely when the pixel is vertical to the edge trend; along the edge, the gray value of the pixel changes smoothly, i.e. along the direction of the edge point, another edge point with small difference between the gray value and the direction can be always found.
Since noise is random, it is difficult for any noise pixel to find another noise point whose gray value and direction differ little along the edge direction.
This embodiment exploits this basic idea to distinguish the actual edge points from the noise points. Specifically, for a specific digital image, after detecting an edge point by using a 4-direction template Sobel operator, if |f (x, y) -f (x+i, y+j) | > TH for any pixel point f (x, y), for any i=0, -1,1; j=0, -1,1 are all true, then the pixel point can be judged as noise point; preferably, the threshold TH has a value of 30.
Subsequently, the noise point f (x, y) is removed from the image. Specifically, the gray value of the point is assigned to 255, and the noise point is changed to a white background.
Through the operation, the color and the background texture of the target object are weakened, and meanwhile, the characteristics of the outline and the shape of the target object are highlighted, so that the phenomenon that the deep learning discrimination rule is over-fitted at the virtual image detail can be avoided to a certain extent.
The resolution of the original inspection image is not changed by the filtering processing, so that the pixel point coordinates of the output target area are the same as the pixel point coordinates in the original image after the inspection image to be detected is subjected to the filtering enhancement processing and then is detected by the detection model based on the deep learning network; therefore, when the detection result is output, the rectangular mark frame of the insulator self-explosion region can be displayed as a layer superimposed in the original image.
3) Insulator self-explosion defect detection by Faster R-CNN deep learning network
The embodiment adopts the fast R-CNN deep learning network widely applied in the deep learning field, and the fast R-CNN deep learning network has remarkable advantages in accuracy and calculation efficiency of target detection.
The Fast R-CNN deep learning network consists of a shared weight layer, a region extraction network (Region proposal network, hereinafter referred to as RPN) and a target detection network (Fast R-CNN, hereinafter referred to as FRCN) which are connected in parallel.
The fast R-CNN deep learning network in the embodiment adopts a 16-layer VGG16 network in a shared weight layer.
The regional extraction network RPN is a full convolutional network with a multiple convolutional layer structure. The RPN provides the initial value of the pixel coordinates of the candidate region to the FRCN network for detection by judging the probability that the pixel belongs to the target to be detected at each pixel position in the final layer of characteristics. The FRCN is further divided into two parallel circumscribed frame regression networks and a target class score network.
Training a fast R-CNN deep learning network by using a small amount of real inspection images, generated virtual inspection images and corresponding defect region marking information, and a training set, and obtaining an insulator self-explosion defect detection model, wherein the detection model is used for insulator self-explosion defect detection and outputs pixel point coordinates of an insulator self-explosion defect target region meeting probability confidence threshold conditions.
After further processing, the target area of the self-explosion defect can be highlighted in the inspection image in a rectangular frame manner.
The insulator self-explosion defect detection process specifically comprises the following steps: inputting a patrol image to be detected, and carrying out pixel-by-pixel convolution operation on the input image layer by adopting a VGG16 network through a fast R-CNN deep learning network sharing weight layer to generate all feature vectors of all pixels in a corresponding layer; inputting the feature vectors into a parallel region extraction network RPN and a target detection network FRCN, comparing the feature vectors of all the pixels of the potential insulator self-explosion target region with the insulator self-explosion defect standard feature vectors generated after training, and carrying out cyclic iteration solution, wherein the matching probability is a confidence score value; and determining the pixel point coordinates of the region of the insulator self-explosion defect larger than a pre-specified threshold as a detected target region. The target region has a corresponding confidence score value.
In specific implementation, the specific steps for detecting the insulator self-explosion defect based on the detection model of the fast R-CNN deep learning network are shown in fig. 3.
3.1 input inspection image generating and extracting target characteristics to be detected
Inputting an original patrol image to a fast R-CNN shared weight layer, and carrying out multi-layer convolution and downsampling treatment on the front 13 layers of a VGG16 network of the shared weight layer; the processed original inspection image can reflect some non-characterized hidden information in the image, and the interference of irrelevant image information is eliminated.
In this embodiment, in the VGG16 network sharing the weight layer, a sliding window with a convolution kernel of 3×3 pixels is set. For each sliding window, a multi-dimensional feature vector may be generated, noted as k; then w×h×k feature vectors are correspondingly generated for each layer of the feature image of the w×h specification of the convolutional neural network.
The sliding window sequentially traverses the original input image and downsamples layer by layer, inputting the acquired feature vectors into the region feature extraction network RPN.
Specifically, different features are extracted by convolution operation, some are color distribution, some are texture features, some are boundary features, some are corner features and the like; this extraction process is blind.
These initially extracted features, after passing through the following convolution layers, will yield more expressive features. The result of the convolution operation can be regarded as a process of calculating the degree of template matching.
Specifically, for the inspection image, 512-dimensional full-connection feature vectors can be generated through a standard VGG16 convolutional neural network.
3.2 depth mining image feature information
Inputting the generated 512-dimensional full-connection feature vector into the full-connection layer of the two branches of the FRCN with the circumscribed frame regression network and the target class score network, integrating the features by the full-connection layer, and delivering the feature to a classifier for classification processing to generate a judgment rule as the detection basis of the subsequent insulator targets.
The advantage of realizing full connection processing by adopting a convolution mode is that the size of an input image is not limited, and the original patrol image with any resolution can be processed.
In a detection model based on a fast R-CNN deep learning network, the two branches of a prediction layer RPN and a judgment layer FRCN respectively have the following functions:
1. RPN: pixel coordinates x and y and width and height w and h of a central point for judging the extraction characteristics of the insulator self-explosion defect target;
2. FRCN: and the method is used for judging that the feature belongs to a non-target background in an insulator target or picture to be identified. The processing mode of sliding window is adopted to ensure that all feature spaces of the two branch-associated convolution layers are traversed, convolution results are input into the regional feature extraction network RPN, and next round of loop iteration is carried out.
The above procedure is implemented by iterative calculations with the following rules.
1. The training area feature extraction network is used for initializing an initial model by adopting an ImageNet pre-training model, and layer parameters shared with the VGG16 network can be directly copied into parameters in the model obtained through ImageNet training;
2. and (3) alternately training the regional characteristic extraction network RPN and the target class score network FRCN to extract the regional characteristics of the insulator self-explosion pixel.
3. The layer loss function uses a gaussian distribution with standard deviation=0.01.
4. And initializing a common convolution layer, performing 1,2 and 3 iterations, performing 50000 iterations, and finishing training.
5. Generating a black box discrimination rule, and generating a black box discrimination rule for each pixel region of an original input image according to the calculation, wherein the black box discrimination rule is specifically the black box characteristic calculation of region pixels, and the confidence value range is from 0 to 1; when the confidence is 1, a perfect match is noted.
In the embodiment, the confidence score is calculated by adopting the Faster R-CNN algorithm source code, and is not changed.
3.3 locating insulator Defect target Pixel region
a. And reducing the dimension of the features acquired by different types of windows to a fixed dimension.
And after the input inspection images to be detected are subjected to convolution operation downsampling, inputting a classification layer, adopting a discrimination rule generated in 3.2, giving confidence scores of objects contained in a sliding window by the classification layer, and discarding a confidence threshold value which is preset and is between 0 and 1, wherein a rectangular pixel area with a high score is used as a positive sample, and a negative sample with a low score is considered.
If the preset confidence coefficient threshold value is 0.9, if the confidence coefficient is more than or equal to 0.9, the detection result is considered to be effective, and the detected target area is further displayed; if the detection result is lower than 0.9, the detection result is considered to be wrong.
b. When the result given by the classification layer judges that the pixel area is larger than the confidence coefficient threshold value, the insulator self-explosion defect target is considered to exist in the pixel area, and the position frame of the target needs to be further subjected to regression correction, so that the rectangular frame area output by the recognition result is accurately attached to the minimum circumscribed rectangle of the insulator self-explosion defect target pixel area.
The overlapping degree judgment threshold value in the position regression correction calculation is calculated by the IOU, and iou=two-frame intersection area/two-frame union area. When the area overlapping degree of the two output pixel areas is larger than 0.3, the two areas are considered to have repeated output, regression calculation is needed to be carried out again, and redundant repeated output is eliminated.
If the overlapping degree of the corresponding insulator self-explosion target pixel region in the input image and the real region of the target in the original input inspection image is more than or equal to a set IOU threshold (for example, 0.5), judging that the pixel region has the insulator target, and setting the region label as 1; if the overlap ratio is less than the IOU threshold (e.g., 0.5), then the region tag is set to 0 and the region pixel is considered to be a non-target background region.
c. And (3) for the region with the label of 1, searching a mapping relation between coordinates of the input image and real target coordinates in the image, finishing a regression positioning process, and determining the pixel coordinate position of the insulator region. The region with a label of 0 is considered to be a false identification and does not participate in subsequent operations.
d. Classifying targets through regression correction, obtaining the coordinates of an accurate boundary frame of the targets in the downsampled image by using a multitask loss function frame regression algorithm, then gradually restoring to the original resolution of the inspection image, outputting the pixel coordinates of the four vertexes of the minimum circumscribed rectangle of the insulator self-explosion defect target area under the pixel coordinate system of the original inspection image, drawing a rectangle and highlighting for manual inspection.
The method for detecting the insulator self-explosion defect based on the inspection image is fully described in stages.
1. Insulator and defect equipment three-dimensional modeling
Measuring the accurate size of small hardware fittings such as a connecting hardware fitting, a buckle and the like by using a vernier caliper, and generating a three-dimensional model by using polygonal modeling;
the laser scanner scans three-dimensional information of various equipment parts of the normal glass insulator string and self-explosion defects of the insulators with four types of defects, and models the three-dimensional information in virtual simulation software.
2. Generating virtual samples
Aiming at the problem that the insulator self-explosion defect sample in the unmanned aerial vehicle inspection image is insufficient, the inspection images are generated in a batch in virtual display software according to the obtained three-dimensional information of the insulator and other parts. The pixel resolution may be arbitrarily set, and an arbitrary resolution between 6000×4000 and 1024×768 is recommended. The virtual image samples contain various insulator self-explosion defect targets generated by programs, the number of insulator defects in each virtual sample inspection image is recommended to be not less than 3, and the normal insulator sheet installation positions between the connecting hardware fittings at the two ends of the insulator string are randomly generated. The marking information of the self-explosion defect of the insulator of the virtual image sample is automatically generated by a program.
3. Manual marking of a small number of real inspection images
The manual marking comprises 500 real inspection images of the insulator self-explosion defects, wherein the number of the insulator self-explosion defects of each type is as balanced as possible, the minimum external rectangle of the target area of the insulator self-explosion defects of each type is marked as the standard target area of the insulator defects in the training set, and marking information is synchronously and correspondingly generated.
4. Forming training set
The training set used for the deep learning of this embodiment is composed of a generated virtual sample and a small amount of real patrol images. The number of training sets is recommended to generate 2000 virtual sample images and 500 real inspection images in batches, and the ratio of the number of virtual samples to the number of real samples is about 4:1. The number can meet the data volume requirement of the deep learning training set, and can balance the accuracy of the recognition of the fast R-CNN model and the cost of manual labeling.
The resolution of the real inspection image covers 6000×4000 to 1024×768 main stream resolution of each existing inspection platform, and is added with line information (such as line name, voltage class, affiliated provincial company, etc.).
5. Filtering enhancement
The virtual and real tour inspection images of the preprocessing training set are enhanced by adopting filtering, the weight of the target geometric outline characteristics in the virtual image and the real image is enhanced as much as possible, and the fitting property of the virtual image and the real image is enhanced.
6. Training Faster R-CNN model
Inputting the training set and the corresponding labeling information into a fast R-CNN deep learning network to train the insulator self-explosion defect detection model.
The invention has been described above with reference to a few embodiments. However, as is well known to those skilled in the art, other embodiments than the above disclosed invention are equally possible within the scope of the invention, as defined by the appended patent claims.
Generally, all terms used in the claims are to be interpreted according to their ordinary meaning in the technical field, unless explicitly defined otherwise therein. All references to "a/an/the [ means, component, etc. ]" are to be interpreted openly as referring to at least one instance of said means, component, etc., unless explicitly stated otherwise. The steps of any method disclosed herein do not have to be performed in the exact order disclosed, unless explicitly stated.
Claims (7)
1. The method for detecting the insulator self-explosion defect based on the inspection image is characterized by comprising the following steps of:
acquiring a plurality of real inspection images shot by an electric line inspection unmanned aerial vehicle and comprising defects of a glass insulator, and manually marking the defects of the insulator self-explosion;
generating a virtual inspection image containing the insulator self-explosion defect and automatically marking the insulator self-explosion defect;
performing filtering enhancement processing on the real patrol image and the virtual patrol image to obtain a real patrol image and a virtual patrol image after filtering enhancement, wherein the filtering enhancement processing comprises the following steps:
1) Traversing the inspection image by adopting four-neighborhood intermediate value filtering to obtain an inspection image after median filtering;
2) Traversing the median filtered inspection image by adopting a Sobel edge operator with 4 direction templates of 0 degree, 45 degrees, 90 degrees and 135 degrees to obtain an edge-protruding inspection image;
3) Denoising the to-be-detected inspection image with the protruding edge by using a four-neighborhood gray scale contrast method to obtain an inspection image after filtering enhancement, wherein the method comprises the following steps:
for any pixel point f (x, y), if |f (x, y) -f (x+i, y+j) | > TH, for any i= -1,1; j= -1,1 are both true, the pixel point is a noise point; wherein the threshold TH is a constant of not less than 20;
Designating a gray value of a pixel point determined as a noise point as 255;
combining the real inspection image and the virtual inspection image after the filtering enhancement according to a preset quantity ratio to form a combined training set;
determining an insulator self-explosion defect detection model based on a fast R-CNN deep learning network by utilizing the combined training set training, wherein when the insulator self-explosion defect detection model determines the self-explosion defect type of the insulator in the real inspection image or the virtual inspection image, the confidence coefficient of the corresponding self-explosion defect type is larger than a preset detection threshold value, if the confidence coefficient of the corresponding self-explosion defect type is larger than the preset detection threshold value, the detection result is determined to be effective, and the area where the insulator self-explosion defect is located is marked by a rectangular frame in the inspection image;
if the confidence coefficient aiming at the corresponding self-explosion defect type is not greater than a preset detection threshold value, determining that the detection result is invalid.
2. The method for detecting an insulator self-explosion defect according to claim 1, wherein,
the self-explosion defect types include: single-chip self-explosion in the glass insulator string, continuous multi-chip self-explosion in the glass insulator string, single-chip self-explosion at the hardware fitting connecting end in the glass insulator string and multi-chip self-explosion at the hardware fitting connecting end of the glass insulator string.
3. The method for detecting an insulator self-explosion defect according to claim 1, further comprising:
acquiring an inspection image to be detected, which is shot by an electric line inspection unmanned aerial vehicle and comprises a glass insulator;
performing filtering enhancement processing on the to-be-detected inspection image to obtain a filtered and enhanced to-be-detected inspection image;
and detecting by using the insulator self-explosion defect detection model, and filtering the reinforced inspection image to be detected to determine whether the inspection image to be detected comprising the glass insulator comprises the self-explosion defect.
4. The method for detecting an insulator self-explosion defect according to claim 1, wherein,
the generating the virtual inspection image containing the insulator self-explosion defect and automatically labeling the insulator self-explosion defect comprises the following steps:
1) Measuring real objects of glass insulators with different self-explosion defect types and normal glass insulators by using a laser scanner to obtain three-dimensional geometric data of the real objects;
2) According to the three-dimensional geometric data, respectively establishing a three-dimensional virtual model aiming at the insulator sheets, the normal insulator sheets and the connecting pieces of different self-explosion defect types, and rendering to obtain defective insulator images, normal insulator images and images of the hardware connecting pieces;
3) Generating an insulator string virtual image according to a preset rule, wherein at least one defective insulator is randomly distributed in different areas of the insulator string virtual image;
4) And simulating and photographing the generated virtual images of the insulator strings from a plurality of different angles, rendering and outputting photographing results into a picture image format, and recording coordinate information of pixel points of the areas where the defective insulators and the adjacent normal insulators are located.
5. The method for detecting an insulator self-explosion defect according to claim 1, wherein,
the manual marking insulator self-explosion defect comprises the following steps:
and manually marking the self-explosion defect areas in the obtained real inspection images of the glass insulators with different self-explosion defect types and the areas where the adjacent normal insulator sheets or the connecting ends of the hardware fittings are located.
6. The method for detecting an insulator self-explosion defect according to claim 1, wherein,
the automatic labeling of the insulator self-explosion defect comprises the following steps:
and determining coordinate information of pixel points of areas where the self-explosion defect insulator sheets and adjacent normal insulator sheets or hardware fitting connecting ends of the self-explosion defect insulator sheets are located in virtual marks of the glass insulators with different self-explosion defect types under a plurality of different shooting visual angles.
7. The method for detecting an insulator self-explosion defect according to claim 3, further comprising:
and after determining that the inspection image to be detected comprising the glass insulator comprises the self-explosion defect, superposing the minimum circumscribed rectangle attached to the insulator sheet with the self-explosion defect and the pixel area where the normal insulators adjacent to the two ends of the insulator sheet are positioned on the inspection image to be detected.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201811376353.6A CN109615611B (en) | 2018-11-19 | 2018-11-19 | Inspection image-based insulator self-explosion defect detection method |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201811376353.6A CN109615611B (en) | 2018-11-19 | 2018-11-19 | Inspection image-based insulator self-explosion defect detection method |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN109615611A CN109615611A (en) | 2019-04-12 |
| CN109615611B true CN109615611B (en) | 2023-06-27 |
Family
ID=66004524
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201811376353.6A Active CN109615611B (en) | 2018-11-19 | 2018-11-19 | Inspection image-based insulator self-explosion defect detection method |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN109615611B (en) |
Families Citing this family (19)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN110148117B (en) * | 2019-04-22 | 2021-07-20 | 南方电网科学研究院有限责任公司 | Power equipment defect identification method and device based on power image and storage medium |
| CN110120091B (en) * | 2019-04-28 | 2023-06-16 | 深圳供电局有限公司 | Method and device for manufacturing electric power inspection image sample and computer equipment |
| CN110147777B (en) * | 2019-05-24 | 2021-03-09 | 合肥工业大学 | Insulator category detection method based on deep migration learning |
| CN110222683A (en) * | 2019-06-11 | 2019-09-10 | 云南电网有限责任公司曲靖供电局 | A kind of quick-fried defect recognition localization method of electric transmission line isolator component based on depth convolutional neural networks |
| CN110503029A (en) * | 2019-08-21 | 2019-11-26 | 云南电网有限责任公司电力科学研究院 | A kind of detection method of transmission line of electricity glass insulator state |
| CN110852988A (en) * | 2019-09-27 | 2020-02-28 | 广东电网有限责任公司清远供电局 | Method, device and equipment for detecting self-explosion of insulator string and storage medium |
| CN110751642A (en) * | 2019-10-18 | 2020-02-04 | 国网黑龙江省电力有限公司大庆供电公司 | Insulator crack detection method and system |
| CN111220619B (en) * | 2019-12-05 | 2023-05-05 | 河海大学常州校区 | Insulator self-explosion detection method |
| CN111182491B (en) * | 2019-12-31 | 2022-03-22 | 淮安中科晶上智能网联研究院有限公司 | Radio frequency tomography-based equipment-free target positioning method and device |
| KR102430090B1 (en) * | 2020-04-20 | 2022-08-11 | 세이지리서치 주식회사 | Computer program, method, and device for generating virtual defect image using artificial intelligence model generated based on user input |
| CN111709918B (en) * | 2020-06-01 | 2023-04-18 | 深圳市深视创新科技有限公司 | Product defect classification method combining multiple channels based on deep learning |
| CN111799697B (en) * | 2020-07-08 | 2021-09-03 | 云南电网有限责任公司电力科学研究院 | Self-explosion simulation method and system for glass insulator of power transmission line |
| CN111784692A (en) * | 2020-08-11 | 2020-10-16 | 国网内蒙古东部电力有限公司 | Method and device for detecting insulator defects in power system and electronic equipment |
| CN112634254A (en) * | 2020-12-29 | 2021-04-09 | 北京市商汤科技开发有限公司 | Insulator defect detection method and related device |
| CN112837315B (en) * | 2021-03-05 | 2023-11-21 | 云南电网有限责任公司电力科学研究院 | Deep learning-based transmission line insulator defect detection method |
| CN113393453A (en) * | 2021-06-28 | 2021-09-14 | 北京百度网讯科技有限公司 | Method, apparatus, device, medium and product for detecting self-bursting insulators |
| CN113469997B (en) * | 2021-07-19 | 2024-02-09 | 京东科技控股股份有限公司 | Method, device, equipment and medium for detecting plane glass |
| CN115063618B (en) * | 2022-08-17 | 2022-11-11 | 成都数之联科技股份有限公司 | Defect positioning method, system, equipment and medium based on template matching |
| CN117541582B (en) * | 2024-01-09 | 2024-04-19 | 山东海纳智能装备科技股份有限公司 | IGBT insulation quality detection method for high-frequency converter |
Citations (5)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN106056088A (en) * | 2016-06-03 | 2016-10-26 | 西安电子科技大学 | Single-sample face recognition method based on self-adaptive virtual sample generation criterion |
| CN107451997A (en) * | 2017-07-31 | 2017-12-08 | 南昌航空大学 | A kind of automatic identifying method of the welding line ultrasonic TOFD D scanning defect types based on deep learning |
| CN107918761A (en) * | 2017-10-19 | 2018-04-17 | 九江学院 | A kind of single sample face recognition method based on multiple manifold kernel discriminant analysis |
| CN108230391A (en) * | 2017-12-13 | 2018-06-29 | 深圳市鸿益达供应链科技有限公司 | Intelligent identification method |
| CN108596886A (en) * | 2018-04-17 | 2018-09-28 | 福州大学 | Aerial Images insulator based on deep learning falls piece fault rapid detecting method |
Family Cites Families (1)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| TWI737659B (en) * | 2015-12-22 | 2021-09-01 | 以色列商應用材料以色列公司 | Method of deep learning - based examination of a semiconductor specimen and system thereof |
-
2018
- 2018-11-19 CN CN201811376353.6A patent/CN109615611B/en active Active
Patent Citations (5)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN106056088A (en) * | 2016-06-03 | 2016-10-26 | 西安电子科技大学 | Single-sample face recognition method based on self-adaptive virtual sample generation criterion |
| CN107451997A (en) * | 2017-07-31 | 2017-12-08 | 南昌航空大学 | A kind of automatic identifying method of the welding line ultrasonic TOFD D scanning defect types based on deep learning |
| CN107918761A (en) * | 2017-10-19 | 2018-04-17 | 九江学院 | A kind of single sample face recognition method based on multiple manifold kernel discriminant analysis |
| CN108230391A (en) * | 2017-12-13 | 2018-06-29 | 深圳市鸿益达供应链科技有限公司 | Intelligent identification method |
| CN108596886A (en) * | 2018-04-17 | 2018-09-28 | 福州大学 | Aerial Images insulator based on deep learning falls piece fault rapid detecting method |
Also Published As
| Publication number | Publication date |
|---|---|
| CN109615611A (en) | 2019-04-12 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN109615611B (en) | Inspection image-based insulator self-explosion defect detection method | |
| CN109598794B (en) | Construction method of three-dimensional GIS dynamic model | |
| Zhu et al. | Concrete column recognition in images and videos | |
| CN111553859A (en) | Laser radar point cloud reflection intensity completion method and system | |
| CN108305260B (en) | Method, device and equipment for detecting angular points in image | |
| CN107818303B (en) | Unmanned aerial vehicle oil and gas pipeline image automatic contrast analysis method, system and software memory | |
| CN113139453A (en) | Orthoimage high-rise building base vector extraction method based on deep learning | |
| CN107590444A (en) | Detection method, device and the storage medium of static-obstacle thing | |
| Wang | Automatic extraction of building outline from high resolution aerial imagery | |
| CN110348307B (en) | Path edge identification method and system for crane metal structure climbing robot | |
| CN115222884A (en) | Space object analysis and modeling optimization method based on artificial intelligence | |
| Sahar et al. | Semiautomatic extraction of building outlines from large-scale aerial images | |
| Guo et al. | Visibility detection approach to road scene foggy images | |
| Adu-Gyamfi et al. | Functional evaluation of pavement condition using a complete vision system | |
| CN113989604A (en) | Tire DOT information identification method based on end-to-end deep learning | |
| CN109658523A (en) | The method for realizing each function operation instruction of vehicle using the application of AR augmented reality | |
| CN112115737B (en) | Vehicle orientation determining method and device and vehicle-mounted terminal | |
| CN116958837A (en) | Municipal facilities fault detection system based on unmanned aerial vehicle | |
| Cheng et al. | Power pole detection based on graph cut | |
| CN113920254B (en) | Monocular RGB (Red Green blue) -based indoor three-dimensional reconstruction method and system thereof | |
| CN113284158B (en) | Image edge extraction method and system based on structural constraint clustering | |
| CN112116561B (en) | Power grid transmission line detection method and device based on image processing fusion network weight | |
| CN113658195A (en) | Image segmentation method and device and electronic equipment | |
| Jiao et al. | Individual building rooftop and tree crown segmentation from high-resolution urban aerial optical images | |
| CN111915632A (en) | Poor texture target object truth value database construction method based on machine learning |
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 | ||
| GR01 | Patent grant | ||
| GR01 | Patent grant |