CN113592822B - Insulator defect positioning method for electric power inspection image - Google Patents
Insulator defect positioning method for electric power inspection image Download PDFInfo
- Publication number
- CN113592822B CN113592822B CN202110880016.6A CN202110880016A CN113592822B CN 113592822 B CN113592822 B CN 113592822B CN 202110880016 A CN202110880016 A CN 202110880016A CN 113592822 B CN113592822 B CN 113592822B
- Authority
- CN
- China
- Prior art keywords
- insulator
- image
- segmentation
- inspection image
- pixel points
- 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 236
- 238000000034 method Methods 0.000 title claims abstract description 73
- 238000007689 inspection Methods 0.000 title claims abstract description 67
- 230000007547 defect Effects 0.000 title claims abstract description 38
- 230000011218 segmentation Effects 0.000 claims abstract description 60
- 238000001514 detection method Methods 0.000 claims abstract description 52
- 238000012549 training Methods 0.000 claims abstract description 24
- 238000013135 deep learning Methods 0.000 claims abstract description 22
- 230000006835 compression Effects 0.000 claims description 17
- 238000007906 compression Methods 0.000 claims description 17
- 230000006870 function Effects 0.000 claims description 16
- 238000002372 labelling Methods 0.000 claims description 14
- 230000008569 process Effects 0.000 claims description 14
- 238000000605 extraction Methods 0.000 claims description 11
- 238000005520 cutting process Methods 0.000 claims description 9
- 238000011176 pooling Methods 0.000 claims description 9
- 238000011478 gradient descent method Methods 0.000 claims description 7
- 230000009466 transformation Effects 0.000 claims description 6
- 230000003993 interaction Effects 0.000 claims description 5
- 238000012545 processing Methods 0.000 claims description 5
- 230000003321 amplification Effects 0.000 claims description 4
- 230000007246 mechanism Effects 0.000 claims description 4
- 238000003199 nucleic acid amplification method Methods 0.000 claims description 4
- 230000009467 reduction Effects 0.000 claims description 4
- 230000003044 adaptive effect Effects 0.000 claims description 3
- 238000004458 analytical method Methods 0.000 claims description 3
- 238000009432 framing Methods 0.000 claims description 3
- 238000013508 migration Methods 0.000 claims description 2
- 230000005012 migration Effects 0.000 claims description 2
- 230000007306 turnover Effects 0.000 claims description 2
- 238000010586 diagram Methods 0.000 description 10
- 230000005540 biological transmission Effects 0.000 description 7
- 230000000694 effects Effects 0.000 description 5
- 230000002950 deficient Effects 0.000 description 4
- 238000011161 development Methods 0.000 description 4
- 230000018109 developmental process Effects 0.000 description 4
- 230000005611 electricity Effects 0.000 description 4
- 238000013527 convolutional neural network Methods 0.000 description 3
- 238000010248 power generation Methods 0.000 description 3
- 230000008901 benefit Effects 0.000 description 2
- 238000013434 data augmentation Methods 0.000 description 2
- 238000009826 distribution Methods 0.000 description 2
- 238000002474 experimental method Methods 0.000 description 2
- 239000000284 extract Substances 0.000 description 2
- 238000003709 image segmentation Methods 0.000 description 2
- 230000006872 improvement Effects 0.000 description 2
- 238000004519 manufacturing process Methods 0.000 description 2
- 238000005457 optimization Methods 0.000 description 2
- 238000011160 research Methods 0.000 description 2
- 230000009286 beneficial effect Effects 0.000 description 1
- 238000007664 blowing Methods 0.000 description 1
- 238000004364 calculation method Methods 0.000 description 1
- 238000012217 deletion Methods 0.000 description 1
- 230000037430 deletion Effects 0.000 description 1
- 238000013461 design Methods 0.000 description 1
- 239000003989 dielectric material Substances 0.000 description 1
- 230000008034 disappearance Effects 0.000 description 1
- 238000010292 electrical insulation Methods 0.000 description 1
- 238000005516 engineering process Methods 0.000 description 1
- 238000009413 insulation Methods 0.000 description 1
- 230000007774 longterm Effects 0.000 description 1
- 238000012423 maintenance Methods 0.000 description 1
- 239000000463 material Substances 0.000 description 1
- 238000012986 modification Methods 0.000 description 1
- 230000004048 modification Effects 0.000 description 1
- 238000010606 normalization Methods 0.000 description 1
- 230000003647 oxidation Effects 0.000 description 1
- 238000007254 oxidation reaction Methods 0.000 description 1
- 238000003672 processing method Methods 0.000 description 1
- 238000012216 screening Methods 0.000 description 1
- 238000013526 transfer learning Methods 0.000 description 1
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
- G06T7/0008—Industrial image inspection checking presence/absence
-
- G—PHYSICS
- G01—MEASURING; TESTING
- G01N—INVESTIGATING OR ANALYSING MATERIALS BY DETERMINING THEIR CHEMICAL OR PHYSICAL PROPERTIES
- G01N21/00—Investigating or analysing materials by the use of optical means, i.e. using sub-millimetre waves, infrared, visible or ultraviolet light
- G01N21/84—Systems specially adapted for particular applications
- G01N21/88—Investigating the presence of flaws or contamination
- G01N21/95—Investigating the presence of flaws or contamination characterised by the material or shape of the object to be examined
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T7/00—Image analysis
- G06T7/10—Segmentation; Edge detection
- G06T7/12—Edge-based segmentation
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T7/00—Image analysis
- G06T7/70—Determining position or orientation of objects or cameras
- G06T7/73—Determining position or orientation of objects or cameras using feature-based methods
-
- 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
- G06T2207/20021—Dividing image into blocks, subimages or windows
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T2207/00—Indexing scheme for image analysis or image enhancement
- G06T2207/20—Special algorithmic details
- G06T2207/20081—Training; Learning
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T2207/00—Indexing scheme for image analysis or image enhancement
- G06T2207/20—Special algorithmic details
- G06T2207/20084—Artificial neural networks [ANN]
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T2207/00—Indexing scheme for image analysis or image enhancement
- G06T2207/20—Special algorithmic details
- G06T2207/20112—Image segmentation details
- G06T2207/20132—Image cropping
-
- 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
-
- 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)
- Computer Vision & Pattern Recognition (AREA)
- Chemical & Material Sciences (AREA)
- Health & Medical Sciences (AREA)
- Life Sciences & Earth Sciences (AREA)
- Quality & Reliability (AREA)
- Analytical Chemistry (AREA)
- Biochemistry (AREA)
- General Health & Medical Sciences (AREA)
- Immunology (AREA)
- Pathology (AREA)
- Image Processing (AREA)
- Image Analysis (AREA)
Abstract
The invention provides an insulator defect positioning method of an electric power inspection image, which comprises the following steps: training a target detection algorithm by using an insulator detection data set to obtain an insulator identification model; training the improved deep learning network U-Net by utilizing the edge segmentation data set to obtain an insulator segmentation model; inputting the inspection image to be processed into an insulator identification model for detection to obtain the number and positions of insulators of the inspection image to be processed, and marking by using a square frame; inputting the marked image into an insulator segmentation model in the second step for segmentation to obtain a binarized insulator image; mathematical modeling is carried out on the insulator strings in the binarized insulator image through a least square method, the insulator defects are detected through the mathematical modeling, and the positions and the number of the insulator string defects are judged. The invention improves the detection precision and the segmentation precision of the insulator, has certain robustness and is not influenced by the size of the insulator in the inspection image.
Description
Technical Field
The invention relates to the technical field of power transmission line inspection, in particular to an insulator defect positioning method of a power inspection image.
Background
With the continuous improvement of social development and productivity, electric energy plays an indispensable role in daily life of people, and the running of large-scale equipment in various aspects of China, society, military, life and the like and the smooth progress of scientific research are not separated from the role of electric energy in the electric energy. Besides traditional thermal power generation, the country strongly supports novel energy sources such as wind power generation, solar power generation and the like to balance the increasing electricity consumption of the country. According to the related data, the electricity generation capacity of China in recent years is the first world, and the China is a large country for electricity generation and electricity utilization.
The electric power system has important roles in the production and life of the national, so that the maintenance of stable and safe operation of the electric power system is very important. Overhead transmission lines are an important part of power systems, and are composed of line poles, wires, insulators, grounding devices and the like. The insulator is a special insulation control, and the main function of the insulator is to realize electrical insulation and mechanical fixation. Therefore, the state of the insulator directly affects the normal operation of the transmission line. Because the insulator is exposed in the air for a long time, the insulator is extremely easy to damage and fall after long-term wind blowing oxidation, and the insulator is also easy to be defective in extreme weather such as storm snow, thunder and lightning weather and the like. In these cases, the insulator may lose its insulating function and its mechanical fixing function, severely threatening the safe operation of the grid. Therefore, power grid staff can regularly patrol insulators on the power transmission line, and safe and stable operation of the power grid is ensured.
The electric power inspection is mainly divided into two modes of manual inspection and unmanned aerial vehicle inspection. The manual inspection requires a worker to operate on a power transmission tower pole, and has the following defects: the difficulty of high-altitude operation is too great, and workers need to climb to electric towers and transmission lines of tens of meters or even tens of meters to carry out inspection, especially in areas with complex geographic environments, and the difficulty of manual inspection is increased; secondly, certain potential safety hazards exist in the overhead operation, and accidents can occur if the workers are improperly protected during the overhead operation; finally, the staff cannot move quickly due to the influence of the high-altitude working place, so that the manual inspection efficiency is low. At present, the power equipment is overhauled mainly by adopting a manual inspection mode in China, but with the development of society and the progress of science and technology, the national power grid is trying to find a new method to replace the traditional manual inspection mode. In recent years, four rotor unmanned aerial vehicles have been used in the field of aerial photography. In the power industry, research is also being conducted on using a four-rotor unmanned aerial vehicle for power inspection. The unmanned aerial vehicle is utilized for electric power inspection to make up for some defects of manual inspection. Firstly, unmanned aerial vehicle can not be limited by the influence of surrounding geographical environment, still can normally work under complicated landform environment, and unmanned aerial vehicle's flexible free characteristics make it can patrol and examine the place that the manual work is difficult to reach, have not only reduced artifical and patrol and examine the risk that probably causes the staff, also greatly reduced the degree of difficulty of patrol and examine, also greatly improved the efficiency of patrol and examine simultaneously.
In the unmanned aerial vehicle power inspection process, a large number of inspection images can be shot, and how to process the inspection images and accurately detect the defects of the insulators is important content to be researched. In recent years, a deep learning-based method has been rapidly developed in the field of image processing, and a deep learning algorithm is used for processing a patrol image, so that a model can learn relevant characteristics of an insulator, and the insulator can be rapidly and efficiently identified and segmented, thereby smoothly completing defect detection. The process does not need human participation, reduces errors caused by human errors, and has important significance for intelligent development of the power grid.
Disclosure of Invention
Aiming at the technical problem that the existing image processing method is not suitable for insulator defect detection, the invention provides an insulator defect positioning method of an electric power inspection image, realizes insulator defect detection based on a deep learning method, and has a certain application value for maintaining safe operation of a power grid and improving the intelligent degree of the power grid.
In order to achieve the above purpose, the technical scheme of the invention is realized as follows: a method for positioning the defects of an insulator of an electric power inspection image comprises the following steps:
step one: labeling the insulator in the acquired inspection image, establishing an insulator detection data set, and training a target detection algorithm by using the insulator detection data set to obtain an insulator identification model;
step two: labeling an insulator in the acquired inspection image, establishing an edge segmentation data set, and training an improved deep learning network U-Net by using the edge segmentation data set to obtain an insulator segmentation model;
step three: inputting the inspection image to be processed into an insulator identification model for detection to obtain the number and positions of insulators of the inspection image to be processed, and marking by using a square frame; inputting the marked image into an insulator segmentation model in the second step for segmentation to obtain a binarized insulator image;
step four: mathematical modeling is carried out on the insulator strings in the binarized insulator image through a least square method, the insulator defects are detected through the mathematical modeling, and the positions and the number of the insulator string defects are judged.
Marking insulators in the inspection image by using LabelImg software, framing all insulators in the inspection image by using a square frame, selecting the category of an object, and establishing an insulator detection data set; the data quantity of the insulator detection data set is increased through a data amplification method of horizontal overturning, random cutting and scaling transformation;
labeling insulators in the inspection image by using Labelme software, extracting edges by using a Canny operator to serve as an edge segmentation dataset, and increasing the data volume of the edge segmentation dataset by adopting a turnover, random cutting and scaling transformation mode;
after labeling of all insulators on a picture is completed, labelme software generates json files, and the json files record information of each labeled insulator; and converting json files of all the inspection images into mask tag files in batches, and performing binarization processing on the mask tag files to obtain an edge segmentation data set.
The target detection algorithm adopts a YOLOv3 algorithm based on a GIoU strategy, and adopts a migration learning and small-batch gradient descent method to train an insulator identification model.
The implementation method of the GIoU strategy comprises the following steps:
1) Calculate the actual frame B g Is defined by the area of:wherein, the actual frame B p Is defined as the boundary frame coordinates of
2) Calculating a prediction frame B p Is defined by the area of:wherein, the prediction frame B p Is +.>And->
3) Calculating a prediction frame B p And actual frame B g Is the intersection I of:
4) Find the minimum closed frame B c Coordinates of (c):
5) Calculating the minimum closed frame B c Is defined by the area of:
6) Calculating the cross-over ratioWherein u=a p +A g -I；
7)
8) Calculating a loss function L IoU ＝1-IoU,L GIOU ＝1-GIoU。
The improved deep learning network U-Net is added with the idea of multi-task learning, and edge extraction tasks are introduced as network branches on the basis of segmentation tasks; an ECA attention mechanism module is added into the improved deep learning network U-Net.
The edge extraction task is equivalent to carrying out deep supervision on the network of the segmentation task, and the feature map finally output by the extraction decoder is used for assisting in edge extraction network branches; the idea of the multi-task learning is to add a branch on the basis of a backbone network to divide edges, the backbone network adopts a classical coding-decoding structure and consists of an encoder and two decoders, the encoder is used for extracting a feature map, and the two decoders are respectively used for extracting edge information and extracting the outline of the whole insulator.
The improved deep learning network U-Net is divided into a compression path and an expansion path, the compression path is composed of 4 blocks, each block uses 3 effective convolutions and 1 maximum pooling downsampling, then the number of feature images obtained after downsampling is multiplied by 2, and a feature image with the size of 32 x 32 is obtained through a series of convolution pooling operations; the expansion path is also composed of 4 blocks, the size of the feature map is multiplied by 2 by deconvolution before each block starts, the number of the feature maps is halved, and then the feature maps are combined with the feature map of the compression path; cutting the feature map of the compression path to obtain the feature map of the expansion path, normalizing the feature map of the compression path after the feature map of the compression path is the same as the feature map of the expansion path, and outputting two feature maps with the same size;
the ECA attention module performs channel-by-channel global average pooling without dimension reduction, capturing local cross-channel interactions by considering each channel and its k neighbors; the size of k adopts an adaptive determination method, and the kernel size k is proportional to the channel dimension.
The improved deep learning network U-Net is optimized by adopting a small batch gradient descent method; in the improved deep learning network U-Net training process, an SGD optimizer is adopted, a Cosine analysis learning rate attenuation strategy is adopted, and when the loss function value is closer to the global minimum value, the learning rate becomes smaller.
The method for detecting the defects of the insulators by utilizing mathematical modeling comprises the following steps:
s1: taking the pixel point at the lower left corner of the binarized insulator image as an origin, taking the horizontal edge as an X axis and the vertical edge as a Y axis, and establishing a rectangular coordinate system; fitting an insulator string in the insulator image by a least square method to obtain a straight line L passing through the center of the insulator string;
s2: obtaining a straight line D which is perpendicular to the straight line L by utilizing a formula that the two straight lines are perpendicular to each other;
s3: scanning on the insulator string by utilizing horizontal movement of the straight line D, and recording the number of pixel points obtained by each scanning; comparing the sum of pixel points on each insulator sheet on the insulator string with a discrimination threshold, and judging that the insulator sheet is normal when the sum of the pixel points is larger than the discrimination threshold; when the sum of the pixel points is smaller than the judging threshold value, judging that the insulator sheet is damaged or missing;
s4: and after all the insulator sheets on the insulator string are scanned, displaying the positions and the number of damaged insulators.
In the step S1, the method for fitting the insulator string in the insulator image by using the least square method includes: traversing all pixel points in the binarized insulator image, storing the abscissa of all white pixel points in an array x, and placing the ordinate of all white pixel points in an array y; fitting all the white pixel points by using a least square method to obtain a center straight line L of the insulator string; let the expression of the straight line equation L be: y=ax+b, where a is the slope of the center line Z and B is the intercept of the center line Z;
the fitting principle of the least square method is to bias the parameters of the straight line L so that the actual value y and the observed value y i The error between the two is the smallest:wherein the coordinates (x j ,y j ) J=0, 1,2, …, N is the number of white pixels;
the bias derivatives of the parameters A and B are obtained respectively:
and (3) finishing to obtain:
the values of the solved parameters A and B are respectively:
data of all observation points (x j ,y j ) J=1, 2..n is substituted into the linear equation of the straight line L to reach the minimum value y=ax+b, and then the best estimated values a and B are obtained to obtain the fitted linear equation of the straight line L.
The method for acquiring the discrimination threshold comprises the following steps: and calculating the sum of pixel points on each insulator sheet according to the pixel points obtained by scanning, and taking the average value as the discrimination threshold value of each binarized insulator image.
The invention has the beneficial effects that: identifying the insulators by using a target detection algorithm, and determining the positions and the number of the insulators; the pure insulator strings are segmented through an image segmentation algorithm; after obtaining a pure insulator string, reading pixel points of the insulator string in an image; fitting pixel points of the insulator string by using a least square method to obtain a straight line L passing through the center of the insulator and a straight line D perpendicular to the straight line L; scanning the straight line D along the straight line L according to the regularity of the insulator string, and reading the number of pixel points scanned each time; the number of pixel points on each insulator sheet is calculated, and the positions of the damage or the deletion of the insulators are obtained by comparing and judging the threshold values, so that the defects of the insulator strings can be effectively detected and positioned.
In the process of insulator detection, the loss function in the YOLO-v3 algorithm is improved, and the GIoU strategy is used for replacing the IoU strategy in the original loss function, so that the insulator detection precision is improved; in the process of dividing the insulator, the U-Net network is improved by combining the thought of multitask learning and an ECA attention mechanism, so that the accuracy of the insulator division is improved.
Drawings
In order to more clearly illustrate the embodiments of the invention or the technical solutions in the prior art, the drawings that are required in the embodiments or the description of the prior art will be briefly described, it being obvious that the drawings in the following description are only some embodiments of the invention, and that other drawings may be obtained according to these drawings without inventive effort for a person skilled in the art.
FIG. 1 is a block diagram of the YOLOv3 algorithm of the present invention.
Fig. 2 shows three cases in which the IoU policy is the same and the GIoU policy is different in the present invention, wherein (a) is IoU =0.33 and giou=0.33, (b) is IoU =0.33 and giou=0.24, and (c) is IoU =0.33 and giou= -0.1.
Fig. 3 is a graph showing comparison of insulator detection results in the present invention, wherein (a) is example 1 and (b) is example 2.
FIG. 4 is a block diagram of U-Net according to the present invention.
Fig. 5 is a schematic diagram of the multi-task learning in the present invention.
Fig. 6 is a block diagram of an ECA attention module in accordance with the present invention.
Fig. 7 is a block diagram of an improved U-Net network in accordance with the present invention.
Fig. 8 shows an exemplary result I of the insulator segmentation in the present invention, where (a) is the result of the direct segmentation of the U-Net algorithm and (b) is the result of the segmentation in the present invention.
Fig. 9 shows an example result II of the insulator segmentation in the present invention, where (a) is the result of the direct segmentation of the U-Net algorithm and (b) is the result of the segmentation in the present invention.
Fig. 10 is a flowchart of insulator defect detection in the present invention.
Detailed Description
The following description of the embodiments of the present invention will be made clearly and completely with reference to the accompanying drawings, in which it is apparent that the embodiments described are only some embodiments of the present invention, but not all embodiments. All other embodiments, which can be made by those skilled in the art based on the embodiments of the invention without any inventive effort, are intended to be within the scope of the invention.
As shown in fig. 1, the method for positioning the defects of the insulators in the power inspection image comprises the following steps:
step one: labeling the insulator in the acquired inspection image, establishing an insulator detection data set, and training a target detection algorithm by using the insulator detection data set to obtain an insulator identification model.
In terms of data, labeling insulators in the inspection image by using LabelImg software: and importing the inspection image into LabelImg software, then framing all insulators in the inspection image by using a square frame, selecting the category of the object, establishing an insulator detection data set, and increasing the data volume of the insulator detection data set by using a data amplification method of horizontal overturn, random cutting and scaling transformation.
The insulator detection dataset is taken as a training dataset, and is a dataset with 1000 power insulator images, which is prepared by taking an insulator picture shot by a Xinjiang M200 power inspection unmanned aerial vehicle, adding the insulator picture collected on the net, and then manually screening. In deep learning, the deeper the network, the more data is needed to support, and only a sufficient amount of data can be used to train a better model. However, in view of reality, the data set of the electric power equipment is not easy to obtain, and the manufacturing process also needs to consume a lot of manpower and material resources, so that while the data set is manufactured in the best effort, some other methods are considered to increase the data amount. Data augmentation is also called data augmentation, which means that limited data is made to produce an equivalent of more data value without substantially increasing the data. In the invention, three modes of horizontal overturn, random clipping and scaling transformation are adopted to amplify the data.
The insulator identification model can realize insulator detection of the inspection image, and the target detection algorithm adopts a YOLOv3 algorithm based on a GIoU strategy. Aiming at the defects of IoU strategy in the YOLOv3 algorithm, the GIoU strategy is used for replacing the defects, the insulator identification model is trained by adopting methods of transfer learning, small-batch gradient descent and the like in the experimental process, the modified algorithm detects insulators which are not detected by the original algorithm in partial pictures, and the recall rate and the detection accuracy are improved to a certain extent.
The whole YOLO-v3 network in the YOLO v3 algorithm is 252 layers in total, the whole network structure is shown in fig. 1, the DBL module is a combination of convolution+bn+leak, and the DBL module is a basic component of the YOLO-v3 network. In resn, n represents a number indicating the number of defective units res-units contained in the defective block res-block. Concat represents tensor stitching, stitching the up-samples of the Darknet-53 middle layer and the later layer.
The insulator position is accurately positioned from the unmanned aerial vehicle inspection image, and the method is a primary task of electric power inspection. Only if the insulator exists in the inspection image, the subsequent segmentation and defect detection can be performed. Therefore, the insulator is positioned by using a target detection algorithm on the basis of deep learning. With the development of deep learning, a target detection method based on a Convolutional Neural Network (CNN) has made a significant breakthrough. The two-stage detection algorithm divides the detection problem into two stages: candidate regions are generated and classified and regressed. Such algorithms require the generation of a large number of candidate regions and are therefore time consuming, but with high accuracy. The single-stage detection algorithm does not need to generate a candidate region, and the detection result can be directly obtained through single detection, so that the detection speed is higher, but the accuracy is not as high as that of the two-stage detection algorithm. Considering timeliness in the unmanned aerial vehicle inspection process, mainly aiming at solving the related work of insulator detection and positioning, a YOLO v3 algorithm with higher detection speed is adopted, and a GIoU strategy is used for replacing a IoU strategy in an original algorithm loss function in consideration of the condition of insufficient accuracy of the YOLO-v3 algorithm.
IoU, which is a full scale intersection ratio, calculates the ratio of the intersection and union of the "predicted bounding box" and the "real bounding box". In the original YOLO-v3 algorithm, the mean square error is used as a loss function, but the mean square error function is sensitive to the scale and cannot reflect the prediction results of different qualities. Therefore, ioU between the predicted box and the real target box is often used as a measure of similarity between the two in a number of related studies of the YOLO-v3 algorithm. IoU is an index commonly used in a target detection task, and the calculation formula of IoU is as follows:
IoU presents the following two problems as a loss function: 1) If the two boxes do not intersect, the distance (overlap) between the two cannot be reflected by definition. Meanwhile, because loss=0, no gradient is returned, and learning and training cannot be performed. 2) IoU cannot accurately reflect the degree of overlap of the two. As shown in fig. 2, three IoU are equal, but their coincidence is seen to be different, where the regression effect is better in fig. 2 (a) and worst in fig. 2 (b). Aiming at the defects of IoU, a new index GIoU is adopted to replace IoU.
The implementation method of the GIoU strategy comprises the following steps:
input: prediction frame B p And actual frame B g Boundary frame coordinates of (c):
and (3) outputting: l (L) IoU ，L GIOU
For prediction block B p Ensure thatAnd->
1) Calculate the actual frame B g Is defined by the area of:
2) Calculating a prediction frame B p Is defined by the area of:
3) Calculating a prediction frame B p And actual frame B g Is the intersection I of:
4) Find the minimum closed frame B c Coordinates of (c):
5) Calculating the minimum closed frame B c Is defined by the area of:
6) Cross-over ratioWherein u=a p +A g -I；
7)
8)L IoU ＝1-IoU,L GIOU ＝1-GIoU。
Optimization is performed using a small batch gradient descent method, which has the following two advantages: on one hand, the data are batched, and the computer only needs to load one batch when updating the gradient each time, so that a large amount of memory is not occupied, and the running speed can be increased; on the other hand, the parameters can be updated in batches, and the direction of the gradient is determined by data of each batch, so that the gradient is not easy to deviate when the gradient is lowered, and the randomness is reduced. In order to make the gradient descent method have better performance, we need to set the learning rate in a reasonable range, and too large learning rate can lead to unstable learning and too small learning rate can lead to too long training time. Setting a reasonable learning rate not only can ensure stable training, but also can reduce training time. Thus, just before the first stage training starts, the learning rate is set to 0.01, and near the second stage training ends, we set the learning rate to 0.0005. After training, identifying and detecting insulators in the aerial image, and determining the positions and the number of the insulators.
Step two: labeling insulators in the acquired inspection images, establishing an edge segmentation data set, and training the improved deep learning network U-Net by using the edge segmentation data set to obtain an insulator segmentation model.
Labeling the inspection image by using Labelme software, extracting edges by using a Canny operator to serve as an edge segmentation data set, and increasing the data quantity of training the edge segmentation data set by adopting data amplification modes such as overturning, random cutting and the like. The inspection image is imported into Labelme software, the outline of the object to be marked is framed, and then the category of the marked object is selected. After labeling of all insulators on a picture is completed, labelme software generates json files, and the information of each labeled insulator is recorded in the json files. And after all the pictures are marked, obtaining a json file corresponding to each picture. The image segmentation task needs that its corresponding label is a. Png/. Bmp file, so by writing a program, json files are converted into mask tag files in batches. After all json files are converted, the mask tag file is subjected to binarization processing, and the edge segmentation data set of the image is manufactured.
According to the characteristics of the segmentation tasks and the thought of multi-task learning, a branch is added to the improved deep learning network U-Net on the basis of the original main trunk network to segment the edge, and an ECA attention mechanism module is added to enable the segmentation result to achieve a better effect.
The segmentation of the insulators in the inspection image is a key step in the defect detection scheme, and the effect of the segmented insulator binarization image directly influences the result of the next defect detection. Insulators are generally present on high-voltage transmission lines, and the complicated geographical environment background of the insulators makes segmentation of the insulators difficult. If the traditional method is adopted, the ideal segmentation effect is difficult to achieve, so that the insulator inspection image is processed by adopting a deep learning segmentation method. U-Net networks are often used for segmentation of medical images, which can achieve good segmentation results with fewer data sets. The inspection image and the medical image of the insulator have certain similarity, the data set of the electric insulator is difficult to collect, the data size is small, and the factors are comprehensively considered.
The U-Net network has no full connection operation and is a classical full convolution network. As shown in fig. 4, the U-Net network can be divided into two parts: a compression path and an expansion path. The compression path is composed of 4 blocks, each block uses 3 effective convolutions and 1 maximum pooling downsampling, then the number of feature images obtained after downsampling is multiplied by 2, and a feature image with the size of 32 x 32 is finally obtained through a series of convolution pooling operations. The extension path is also composed of 4 blocks, each block is multiplied by 2 by deconvolution before starting, and the number of blocks is halved (except for the last layer), and then the blocks are combined with the feature diagram of the compression path which is symmetrical on the left side. However, the feature patterns of the left side compression path and the right side expansion path are different in size and cannot be directly merged. Therefore, the U-Net network performs normalization after cutting the characteristic diagram of the compression path to obtain the same size as the characteristic diagram of the expansion path. Through the above operation, two feature maps of size 388 x 388 are finally output.
In the process of insulator segmentation, if a U-Net network is directly adopted for segmentation, the whole contour of the insulator can be segmented out according to the obtained result, but the edge segmentation of the insulator has the condition of unclear edges. In order to better solve the problem of incomplete edge of the electric tower insulator segmentation, the invention adopts the thought of multi-task learning, and introduces an edge extraction task as a network branch on the basis of segmentation tasks, as shown in fig. 5. Meanwhile, as the two tasks are quite similar, the edge extraction task is equivalent to the deep supervision of the network of the split tasks, and the problems of gradient disappearance and the like can be alleviated. The invention uses the thought of multitask learning in the U-Net segmentation network, and extracts the characteristic diagram finally output by the decoder for assisting the edge extraction network branch so as to help the network to better extract the characteristics and pay more attention to the acquisition of the edge information of the insulator. The backbone part of the network adopts a classical coding-decoding structure, is similar to the design of U-Net, and consists of an encoder and two decoders, wherein the encoder is used for extracting the characteristic diagram, and the two decoders are respectively used for extracting the edge information and the whole insulator contour.
As shown in fig. 7, the present invention introduces an ECA attention module (an effective channel attention module for deep CNN), which not only can effectively perform cross-channel interaction, but also avoids dimension reduction. After channel-by-channel global average pooling without dimension reduction, the ECA attention module captures the local cross-channel interactions by considering each channel and its k neighbors. ECA can be implemented by a fast one-dimensional convolution of size k, where k is the size of the kernel, representing the coverage of local cross-channel interactions, i.e. how many neighbors participate in the attention prediction of a channel, as shown in fig. 6. The size of k employs an adaptive determination method, where the kernel size k is proportional to the channel dimension.
In the training strategy, the optimization is still carried out by adopting a small batch gradient descent method, so that the error fluctuation during training is in a reasonable range, the accuracy of a final result is ensured, the training process is quickened, and the memory of a computer is saved, so that the batch size of each training is selected to be 8 after multiple attempts. The momentum parameter value is empirically set to 0.9 using an SGD optimizer. By adopting the Cosine analysis learning rate attenuation strategy, the learning rate becomes smaller and smaller when the loss function value is closer and closer to the global minimum value in the gradient descending process, and the initial learning rate is set to be 3e-3. Finally, through experiments, after iteration to 150 epochs, the loss function tends to be stable, and training is completed. Finally, by comparing with the traditional segmentation method and the original U-Net network algorithm, the comparison experiment result shows that the trained insulator segmentation model has good performance, the segmented insulator string has clear outline, and the segmentation result is better than the direct segmentation result of the original algorithm.
Step three: inputting the inspection image to be processed into an insulator identification model for detection to obtain the number and positions of insulators of the inspection image to be processed, and marking by using a square frame; and (3) inputting the marked image into an insulator segmentation model in the second step for segmentation to obtain a binarized insulator image.
As shown in fig. 3, the result of the insulator detection by the insulator identification model of the invention can be seen from fig. 8 and 9, the improved algorithm of the invention is obviously better than the original algorithm in the segmentation effect, and the edge segmentation of the insulator is clearer.
Step four: mathematical modeling is carried out on the insulator strings in the binarized insulator image through a least square method, the insulator defects are detected through the mathematical modeling, and the positions and the number of the insulator string defects are judged.
The pixel point distribution of the insulator image has certain regularity, and according to the regularity, modeling is firstly carried out on the insulator string by using a least square method, and a straight line passing through the center point of the insulator is fitted; and secondly, linear scanning the insulator string, calculating pixel points of each insulator sheet according to the distribution rule of the pixel points, and comparing the pixel points with a set threshold value to judge whether the insulator sheet has defects. The modeling method has the greatest advantages of having certain robustness and being not influenced by the size of the target object in the inspection image.
The binarized insulator image is a pure insulator string, after the pure insulator string is obtained, the coordinates of pixel points of the insulator string in the image are read, least square fitting is carried out on the insulator string, a straight line L passing through the center of the insulator string is obtained, and meanwhile, a formula with two mutually perpendicular straight lines is utilized, so that a straight line D perpendicular to the straight line L is obtained; scanning on the insulator string by using a straight line D, and recording the number of pixel points obtained by each scanning; according to certain regularity of the appearance of the insulator string, the sum of pixel points on each insulator sheet can be calculated, and the average value is taken as a discrimination threshold; comparing the sum of pixel points on each insulator sheet with a judging threshold value, and judging that the insulator is damaged or defective if the sum of pixel points on each insulator sheet is smaller than the judging threshold value; if the value is larger than the judging threshold value, judging that the insulator is normal. The insulator string is provided with a plurality of insulator sheets, and the insulator sheets are ellipsoids with thick middle and thin two ends.
As shown in fig. 10, the method for detecting the insulator defect by using mathematical modeling is as follows:
s1: the pixel point at the lower left corner of the segmented image is taken as an origin, the horizontal edge is taken as an X axis, the vertical edge is taken as a Y axis, and a rectangular coordinate system is established; fitting the insulator string by a least square method to obtain a straight line L passing through the center of the insulator string.
The shape of the insulator string in the inspection image shows that the insulator string is linear, all white pixels are distributed regularly, and the insulator string can be fitted through a least square method by utilizing the regularity, so that a straight line L passing through the center of the insulator string is fitted. The least square method is a commonly used data fitting method by minimizing the sum of squares of errors and finding the best function match of the data.
In the binarized insulator image, the insulator string is a white pixel point (with a value of 255), other background elements are black pixel points (with a value of 0), and the insulator string and the background are completely separated. And establishing a rectangular coordinate system by taking the pixel point at the lower left corner of the image as an original point, taking the horizontal edge as an X axis and the vertical edge as a Y axis. Traversing all the pixels, then storing the abscissa of all the white pixels in an x-array, and storing the ordinate of all the white pixels in a y-array. After all the pixel points are traversed, fitting all the white pixel points by using a least square method to obtain a center straight line L of the insulator string. First, assume that the expression of the linear equation L is:
Y＝AX+B
wherein A is the slope of the center line L, and B is the intercept of the center line L. According to the coordinates (x i ,y i ) I=0, 1,2, …, N is the number of white pixels, and the parameters a and B of the center line L can be obtained. The fitting principle of the least square method is to determine the deviation of the parameters of the straight line so as to lead the actual value y and the observed value y i The error between the two is the smallest:
the bias derivatives for parameters A and B in the formula can be obtained:
the above formula is arranged to obtain:
by solving the above equation, the values of parameters A and B can be obtained:
the principle of the least square method for fitting a straight line is to fit the data (x j ,y j ) J=1, 2..n is substituted into the linear equation of the line L to be the minimum value, and then the best estimated value is obtainedAnd->And a fitted straight line L equation can be obtained. And drawing a straight line L in the binary image according to an equation of the straight line L.
S2: and (3) obtaining a straight line D which is perpendicular to the straight line L by using a mathematical formula that the two straight lines are perpendicular to each other, scanning the insulator string by using the straight line D, and recording the number of the pixel points obtained by each scanning.
Finding a straight line D perpendicular to the straight line L, and finding the starting point and the end point of the insulator string according to the number of the pixel points on the straight line D. In the rectangular coordinate system of the previous step, the straight line D is perpendicular to the straight line L, and the slope of the straight line L is known, so that the slope of the straight line D can be obtained.
S3: and in the process of scanning the straight line D on the insulator string, recording the number of the pixel points obtained in each scanning process. When one insulator sheet is scanned, counting and summing the number of pixel points on the insulator sheet, comparing the counted number of pixel points with a calculated judging threshold value, and judging that the insulator sheet is normal when the number of pixel points is larger than the judging threshold value; and when the number of the pixel points is smaller than the judging threshold value, judging that the insulator sheet is damaged or missing.
S4: finally, after the insulator string is scanned, the positions and the numbers of the broken insulators are displayed.
The foregoing description of the preferred embodiments of the invention is not intended to be limiting, but rather is intended to cover all modifications, equivalents, alternatives, and improvements that fall within the spirit and scope of the invention.
Claims (5)
1. The method for positioning the defects of the insulators of the power inspection image is characterized by comprising the following steps of:
step one: labeling the insulator in the acquired inspection image, establishing an insulator detection data set, and training a target detection algorithm by using the insulator detection data set to obtain an insulator identification model;
the target detection algorithm adopts a YOLOv3 algorithm based on a GIoU strategy, and adopts a migration learning and small-batch gradient descent method to train an insulator identification model;
step two: labeling an insulator in the acquired inspection image, establishing an edge segmentation data set, and training an improved deep learning network U-Net by using the edge segmentation data set to obtain an insulator segmentation model;
labeling insulators in the inspection image by using Labelme software, extracting edges by using a Canny operator to serve as an edge segmentation dataset, and increasing the data volume of the edge segmentation dataset by adopting a turnover, random cutting and scaling transformation mode;
after labeling of all insulators on a picture is completed, labelme software generates json files, and the json files record information of each labeled insulator; converting json files of all inspection images into mask tag files in batches, and performing binarization processing on the mask tag files to obtain an edge segmentation data set;
the improved deep learning network U-Net is added with the idea of multi-task learning, and edge extraction tasks are introduced as network branches on the basis of segmentation tasks; an ECA attention mechanism module is added into the improved deep learning network U-Net;
the edge extraction task is equivalent to carrying out deep supervision on the network of the segmentation task, and the feature map finally output by the extraction decoder is used for assisting in edge extraction network branches; the idea of the multi-task learning is to add a branch on the basis of a backbone network to divide edges, wherein the backbone network adopts a classical coding-decoding structure and consists of an encoder and two decoders, the encoder is used for extracting a feature map, and the two decoders are respectively used for extracting edge information and extracting the outline of the whole insulator;
the improved deep learning network U-Net is divided into a compression path and an expansion path, the compression path is composed of 4 blocks, each block uses 3 effective convolutions and 1 maximum pooling downsampling, then the number of feature images obtained after downsampling is multiplied by 2, and a feature image with the size of 32 x 32 is obtained through a series of convolution pooling operations; the expansion path is also composed of 4 blocks, the size of the feature map is multiplied by 2 by deconvolution before each block starts, the number of the feature maps is halved, and then the feature maps are combined with the feature map of the compression path; cutting the feature map of the compression path to obtain the feature map of the expansion path, normalizing the feature map of the compression path after the feature map of the compression path is the same as the feature map of the expansion path, and outputting two feature maps with the same size;
the ECA attention module performs channel-by-channel global average pooling without dimension reduction, capturing local cross-channel interactions by considering each channel and its k neighbors; the size of k adopts an adaptive determination method, and the size of the kernel k is in direct proportion to the dimension of the channel;
the improved deep learning network U-Net is optimized by adopting a small batch gradient descent method; in the improved deep learning network U-Net training process, an SGD optimizer is adopted, a Cosine analysis learning rate attenuation strategy is adopted, and when the loss function value is more and more close to the global minimum value, the learning rate becomes smaller and smaller;
step three: inputting the inspection image to be processed into an insulator identification model for detection to obtain the number and positions of insulators of the inspection image to be processed, and marking by using a square frame; inputting the marked image into an insulator segmentation model in the second step for segmentation to obtain a binarized insulator image;
step four: mathematical modeling is carried out on the insulator strings in the binarized insulator image through a least square method, the insulator defects are detected through the mathematical modeling, and the positions and the number of the insulator string defects are judged;
the method for detecting the defects of the insulators by utilizing mathematical modeling comprises the following steps:
s1: taking the pixel point at the lower left corner of the binarized insulator image as an origin, taking the horizontal edge as an X axis and the vertical edge as a Y axis, and establishing a rectangular coordinate system; fitting an insulator string in the insulator image by a least square method to obtain a straight line L passing through the center of the insulator string;
s2: obtaining a straight line D which is perpendicular to the straight line L by utilizing a formula that the two straight lines are perpendicular to each other;
s3: scanning on the insulator string by utilizing horizontal movement of the straight line D, and recording the number of pixel points obtained by each scanning; comparing the sum of pixel points on each insulator sheet on the insulator string with a discrimination threshold, and judging that the insulator sheet is normal when the sum of the pixel points is larger than the discrimination threshold; when the sum of the pixel points is smaller than the judging threshold value, judging that the insulator sheet is damaged or missing;
s4: and after all the insulator sheets on the insulator string are scanned, displaying the positions and the number of damaged insulators.
2. The method for positioning the insulator defect of the electric power inspection image according to claim 1, wherein in the first step, labeling insulators in the inspection image by using LabelImg software, framing all insulators in the inspection image by using square frames, and selecting the category to which an object belongs to establish an insulator detection data set; and the data quantity of the insulator detection data set is increased through a data amplification method of horizontal overturning, random cutting and scaling transformation.
3. The method for positioning the insulator defect of the power inspection image according to claim 1 or 2, wherein the implementation method of the GIoU strategy is as follows:
1) Calculate the actual frame B g Is defined by the area of:wherein, the actual frame B g Is defined as the boundary frame coordinates of
2) Calculating a prediction frame B p Is defined by the area of:wherein, the prediction frame B p Is defined as the boundary frame coordinates ofAnd->
3) Calculating a prediction frame B p And actual frame B g Is the intersection I of:
4) Find the minimum closed frame B c Coordinates of (c):
5) Calculating the minimum closed frame B c Is defined by the area of:
6) Calculating the cross-over ratioWherein u=a p +A g -I；
7)
8) Calculating a loss function L IoU ＝1-IoU,L GIOU ＝1-GIoU。
4. The method for positioning the insulator defect of the power inspection image according to claim 1, wherein the method for fitting the insulator string in the insulator image by the least square method in the step S1 is as follows: traversing all pixel points in the binarized insulator image, storing the abscissa of all white pixel points in an array x, and placing the ordinate of all white pixel points in an array y; fitting all the white pixel points by using a least square method to obtain a center straight line L of the insulator string; let the expression of the straight line equation L be: y=ax+b, where a is the slope of the center line Z and B is the intercept of the center line Z;
the fitting principle of the least square method is to bias the parameters of the straight line L so that the error between the actual value y and the observed value y' is minimum:wherein the coordinates (x j ,y j ) J=0, 1,2, …, N is the number of white pixels;
the bias derivatives of the parameters A and B are obtained respectively:
and (3) finishing to obtain:
the values of the solved parameters A and B are respectively:
data of all observation points (x j ,y j ) J=1, 2..n is substituted into the linear equation of the line L such thatWhen the minimum value Y=AX+B is reached, the best estimated values A and B are obtained, and a fitted straight line L equation is obtained.
5. The method for locating an insulator defect of a power inspection image according to claim 4, wherein the method for acquiring the discrimination threshold is as follows: and calculating the sum of pixel points on each insulator sheet according to the pixel points obtained by scanning, and taking the average value as the discrimination threshold value of each binarized insulator image.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202110880016.6A CN113592822B (en) | 2021-08-02 | 2021-08-02 | Insulator defect positioning method for electric power inspection image |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202110880016.6A CN113592822B (en) | 2021-08-02 | 2021-08-02 | Insulator defect positioning method for electric power inspection image |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN113592822A CN113592822A (en) | 2021-11-02 |
| CN113592822B true CN113592822B (en) | 2024-02-09 |
Family
ID=78253620
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202110880016.6A Active CN113592822B (en) | 2021-08-02 | 2021-08-02 | Insulator defect positioning method for electric power inspection image |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN113592822B (en) |
Families Citing this family (6)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN114061476B (en) * | 2021-11-17 | 2023-04-18 | 国网宁夏电力有限公司建设分公司 | Method for detecting deflection of insulator of power transmission line |
| CN115082695A (en) * | 2022-05-31 | 2022-09-20 | 中国科学院沈阳自动化研究所 | Transformer substation insulator string modeling and detecting method based on improved Yolov5 |
| CN115115590B (en) * | 2022-06-23 | 2024-03-08 | 华南理工大学 | Composite insulator overheat defect detection method based on rotary RetinaNet |
| CN115272310B (en) * | 2022-09-26 | 2023-01-03 | 江苏智云天工科技有限公司 | Method and device for detecting defects of workpiece |
| CN116152211A (en) * | 2023-02-28 | 2023-05-23 | 哈尔滨市科佳通用机电股份有限公司 | Identification method for brake shoe abrasion overrun fault |
| CN116468730B (en) * | 2023-06-20 | 2023-09-05 | 齐鲁工业大学(山东省科学院) | Aerial Insulator Image Defect Detection Method Based on YOLOv5 Algorithm |
Citations (4)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN109118479A (en) * | 2018-07-26 | 2019-01-01 | 中睿能源（北京）有限公司 | Defects of insulator identification positioning device and method based on capsule network |
| CN112184746A (en) * | 2020-08-27 | 2021-01-05 | 西北工业大学 | Transmission line insulator defect analysis method |
| CN112233092A (en) * | 2020-10-16 | 2021-01-15 | 广东技术师范大学 | Deep learning method for intelligent defect detection of unmanned aerial vehicle power inspection |
| CN112508973A (en) * | 2020-10-19 | 2021-03-16 | 杭州电子科技大学 | MRI image segmentation method based on deep learning |
Family Cites Families (1)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN111259809B (en) * | 2020-01-17 | 2021-08-17 | 五邑大学 | Unmanned aerial vehicle coastline floating garbage inspection system based on DANet |
-
2021
- 2021-08-02 CN CN202110880016.6A patent/CN113592822B/en active Active
Patent Citations (4)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN109118479A (en) * | 2018-07-26 | 2019-01-01 | 中睿能源（北京）有限公司 | Defects of insulator identification positioning device and method based on capsule network |
| CN112184746A (en) * | 2020-08-27 | 2021-01-05 | 西北工业大学 | Transmission line insulator defect analysis method |
| CN112233092A (en) * | 2020-10-16 | 2021-01-15 | 广东技术师范大学 | Deep learning method for intelligent defect detection of unmanned aerial vehicle power inspection |
| CN112508973A (en) * | 2020-10-19 | 2021-03-16 | 杭州电子科技大学 | MRI image segmentation method based on deep learning |
Non-Patent Citations (4)
| Title |
|---|
| Hamid Rezatofighi et al..Generalized Intersection over Union: A Metric and A Loss for Bounding Box Regression.《2019 IEEE/CVF Conference on Computer Vision and Pattern Recognition》.2020,第658-666页. * |
| Lina Yao，Qin Yaoyao.Insulator Detection Dased on GIOU-YOLOv3.《2020 Chinese Automation Congress (CAC)》.2021,第Ⅱ节，第Ⅳ—Ⅵ节. * |
| 袁金丽 等.改进U 型残差网络用于肺结节检测.《计算机工程与应用https://kns.cnki.net/kcms/detail/11.2127.TP.20210330.1429.008.html》.2021,第1节第1段，第1.1.2节，第1.3节，第2节第2段. * |
| 陈文浩 等.无人机电网巡检中的绝缘子缺陷检测与定位.《计算机应用》.2019,第39卷(第S1期),第5节. * |
Also Published As
| Publication number | Publication date |
|---|---|
| CN113592822A (en) | 2021-11-02 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN113592822B (en) | Insulator defect positioning method for electric power inspection image | |
| CN108961235B (en) | Defective insulator identification method based on YOLOv3 network and particle filter algorithm | |
| Wang et al. | High-voltage power transmission tower detection based on faster R-CNN and YOLO-V3 | |
| CN112070135A (en) | Power equipment image detection method and device, power equipment and storage medium | |
| CN111046950B (en) | Image processing method and device, storage medium and electronic device | |
| CN112926556B (en) | Semantic segmentation-based aerial photography transmission line broken strand identification method and system | |
| CN111814597A (en) | Urban function partitioning method coupling multi-label classification network and YOLO | |
| CN111859805A (en) | Method for detecting topological relation of power drawing based on artificial intelligence | |
| CN116206112A (en) | Remote sensing image semantic segmentation method based on multi-scale feature fusion and SAM | |
| CN113902792A (en) | Building height detection method and system based on improved RetinaNet network and electronic equipment | |
| CN114882373A (en) | Multi-feature fusion sandstorm prediction method based on deep neural network | |
| CN113191419B (en) | Sag homologous event detection and type identification method based on track key point matching and region division | |
| CN113536944A (en) | Distribution line inspection data identification and analysis method based on image identification | |
| CN116580285B (en) | Railway insulator night target identification and detection method | |
| CN113034511A (en) | Rural building identification algorithm based on high-resolution remote sensing image and deep learning | |
| CN116452604A (en) | Complex substation scene segmentation method, device and storage medium | |
| CN116385950A (en) | Electric power line hidden danger target detection method under small sample condition | |
| CN116403071A (en) | Method and device for detecting few-sample concrete defects based on feature reconstruction | |
| Hao et al. | Based on Surf feature extraction and insulator damage identification for capsule networks | |
| Li et al. | Automated bridge crack detection based on improving encoder–decoder network and strip pooling | |
| CN114418968A (en) | Power transmission line small target defect detection method based on deep learning | |
| CN113205543A (en) | Laser radar point cloud trunk extraction method based on machine learning | |
| CN114187501A (en) | Package detection method, device and system | |
| CN112465072A (en) | Excavator image identification method based on YOLOv4 model | |
| CN117807518B (en) | Automatic identification method, system and equipment for slot line or shear line on conventional weather diagram |
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 |