《大数据管理概念技术与挑战 (2).pdf》由会员分享，可在线阅读，更多相关《大数据管理概念技术与挑战 (2).pdf（8页珍藏版）》请在taowenge.com淘文阁网|工程机械CAD图纸|机械工程制图|CAD装配图下载|SolidWorks_CaTia_CAD_UG_PROE_设计图分享下载上搜索。
1、 Localization of Multiple Insulators by Orientation Angle Detection and Binary Shape Prior Knowledge ABSTRACTFailure of insulators leads to failure of transmission system and causes heavy loss to the power sector.This has necessitated helicopter inspection on transmission line.Therefore,localizing i
2、nsulators in collected aerial image is an essential step for automatic fault detection.This paper proposes a localization method of multiple insulators with different angles in complex aerial image based on Orientation Angle Detection and Binary Shape Prior Knowledge(OAD-BSPK).Detect possible orient
3、ation angles of insulators preliminarily.For each possible angle,reserve insulator by its binary shape prior knowledge.Traverse all the possible angles,and multiple insulators can be localized in aerial image finally.This paper uses a large number of real aerial captured images as experimental image
4、s.The results show that this method can localize multiple insulators with different orientation angles in complex aerial image,and has higher positioning precision,lower computational complexity and lower time consuming compared with existing methods.This method can meet the requirements of insulato
5、rs real-time and accurate localization.Index Terms-Multiple insulators,localization,aerial image,orientation angle,binary shape prior knowledge.1 INTRODUCTION THE insulator is indispensable equipment in the power system with the dual function of electrical insulation and mechanical support.And its e
6、xternal failure such as crack,surface pollution and damage may harm the safe operation resulting in huge economic losses and casualties in power transmission system 1,2.Therefore,monitoring status of insulator has an important practical significance 3.The helicopter inspection is an effective patrol
7、 way on transmission lines.Using aerial image for condition monitoring of insulator is the development direction of insulators fault diagnosis technology.So it is prerequisite to localize insulators in aerial images.There are some methods of the localization of target.Multi-Threshold OTSU(MT-OTSU)4-
8、6 has high computation complexity and bad robust leading to hardly localize target in real time in literature 4.Literatures 7,8 propose novel texture segmentation algorithm based on active contour model for homogeneous and inhomogeneous aerial insulator images.They can obtain good localization effec
9、ts,but the background of test image is relatively simple and the texture feature of impurity has a high contrast with texture feature of insulator.These methods need to set initial contour,which impairs their real-time,and may not obtain exact localization results when dealing with multiple insulato
10、rs.Literature 9 applies Pulse Coupled Neural Network(PCNN)to segment insulators image and inhibit background to some extent,and coarsely localizes insulator by Generalized Hough Transform(GHT)10-12 using single insulator as template.GHT can realize the segmentation of images with multiple targets,bu
11、t the calculation of algorithm increases quickly with the growth of the dimensions of parameter space.Literature 13 proposes an insulator detection method using Adaptive Neuro-Fuzzy Inference System(ANFIS).Obtain the required cluster by applying k-means clustering,draw bounding box surrounding those
12、 pixels whose intensity row-wise,column-wise and diagonal-wise are appropriate,and get the locations of the insulator in the image by ANFIS.This method can realize localization of multiple insulators in relatively simple background,but it fails to address the location of insulators in complex backgr
13、ound.Literature 14 uses Support Vector Manuscript received on 28 May 2014,in final form 5 June 2015,accepted 30 June 2015.IEEE Transactions on Dielectrics and Electrical Insulation Vol.22,No.6;December 20153421DOI:10.1109/TDEI.2015.004741 Machines(SVM)instead of ANFIS to realize the localization of
14、insulators in complex background.Literature 15 adopts Hough Transform and SVM to extract insulators from the given image.But SVM needs many datasets as input to successfully train SVM,which may increase the computation and complexity of the localization of insulators.Literature 13-15 can realize the
15、 localization of multiple insulators in vertical direction or with small angle inclination.But for multiple insulators with different wide-angle deflection,the intensity of bounding box covering insulator and other targets would reduce,which would result in the loss of insulators.The above localizat
16、ion methods have high computation and complexity when dealing with multiple insulators in complex background,and may be helpless when there are multiple insulators with different orientation angles in aerial image.This paper finds that the whole string of insulator has obvious shape prior knowledge
17、compared with other targets in its binary image,which can be used to greatly improve the positioning accuracy and efficiency,even though there are a lot of insulator types and different shape of single insulator on transmission line.The paper proposes a novel localization method of multiple insulato
18、rs based on Orientation Angle Detection and Binary Shape Prior Knowledge(OAD-BSPK)that can localize insulators in complicated aerial images.This method firstly detects orientation angles of insulators and obtains the possible orientation preliminarily.Then for each possible orientation angle,reserve
19、 insulator by its binary shape prior knowledge,and localize insulator by minimum circumscribed rectangular box.Traverse all the possible angles,and the multiple insulators can be localized in aerial image finally.This method is succinct,and has high positioning accuracy,low computational complexity
20、and low time consuming for the localization of multiple insulators with different orientation angles in complex aerial image.This method will lay the foundation for real-time automatic fault detection of insulators.2 PROCEDURE OF THE PROPOSED METHOD This paper presents a novel methodology for the lo
21、calization of multiple insulators with different orientation angles in complex aerial image.The localization flow chart of insulator is shown in Figure 1.Figure 1.The localization flow chart of insulator.3422Z.Zhao et al.:Localization of Multiple Insulators by Orientation Angle Detection 2.1 IMAGE P
22、REPROCESSING 2.1.1 IMAGE BINARIZATION Convert the original image into binary image by threshold segmentation to realize the separation of foreground and background.2.1.2 MORPHOLOGICAL FILTER There are a lot of internal holes in the image,and the edges are blurry,which lead to difficult edge extracti
23、on.This paper adopts double cascade filtering method to erode and dilate the binary image.And after morphological filtering,internal holes would be filled up,and the edges become smooth and clear.2.1.3 THE REMOVAL OF RESIDUAL REGIONS It still exists in small residual regions which may be incomplete
24、electric wire,tower or partial vegetation after filtering.Set a threshold based on human experiences and remove the regions whose areas are less than the value.The step of image preprocessing can obtain clear binary image and can improve detection accuracy of orientation angle of insulator.2.2 THE D
25、ETECTION OF POSSIBLE ORIENTATION ANGLES OF INSULATORS 2.2.1 EXTRACTION OF EDGE CONTOURS Extract images contours after preprocessing,and link edge pixels together into lists of sequential edge points.A contour starts/stops at an ending or a junction with another contour.And the contour would be disca
26、rded if it is less than a specific value.2.2.2 EDGE SHARPENING The contour of binary image is concave-convex and irregular curve,which brings a lot of trouble for the orientation angle detection of insulator.This paper uses a large number of straight line segments with specified tolerance to approxi
27、mate the irregular curve contour.Thus each contour is composed by plenty straight line segments and the points in each contour can be formed as a list of connected edge points.2.2.3 THE DEFINITION OF ORIENTATION ANGLE For a list of connected edge points,define the orientation angle for each point,an
28、d all the angles are composed of a set of sequential angular variation.The points with sign-Changing angle would be extracted as candidate points.If the number of candidate points of the contour is too less,the contour can be ignored as a pseudo-insulator.As shown in Figure 2,a,b,c are three adjacen
29、t edge points in the list.Define the angle of vertical direction is 0,then the angle of a is-45,and the angle of b and the angle of c are 45,thus b is the points with sign-Changing angle which can be defined as a candidate point.Figure 2.The schematic of orientation angles of edge points.2.2.4 THE E
30、XTRACTION OF ACCURATE POINTS BY RANDOM SAMPLE CONSENSUS(RANSAC)For the extracted candidate points including error points in a list,the accurate points can be picked up by RANSAC 16,17,and the straight line which the points locate in can be determined as the possible direction orientation of insulato
31、r preliminarily.After all lists are traversed,all the possible orientation angles of insulators can be obtained,which can effectively solve the problem of detection of multiple insulators with different orientation angles in aerial image.2.3 THE RESERVASION OF INSULATORS BY BINARY SHAPE PRIOR KNOWLE
32、DGE For a possible orientation angle,rotate binary aerial image with the angle at opposite direction to obtain image with insulator in vertical direction,convert binary image into digital description,and obtain the location of insulator by its binary shape prior knowledge.2.3.1 THE FIRST BINARY SHAP
33、E PRIOR KNOWLEDGE Set the target value to 1 and background value to 0,extract each column of the image respectively,and obtain n groups of 0-1 sequences l1,l2,lj,ln where n is column of the image.Insulator is composed of umbrella plates whose digital sequences are 0s and 1s alternately.The thickness
34、 of each umbrella plate is almost equal,which means that the number of each consecutive 1s in sequence lj is similar.Firstly,search the pixel points changed from 1 to 0 which are edge points of general insulators umbrella plates in sequence lj,and mark their positions.Secondly,count the number of co
35、nsecutive 1s before each edge point as O1,O2,Oj,Os composed of O where s is the number of edge points.The point can be defined and kept as true edge point if the sequence meets the conditions of equation(1)where Nmin is the least number of umbrella plates;otherwise the points are IEEE Transactions o
36、n Dielectrics and Electrical Insulation Vol.22,No.6;December 20153423 pseudo edge points which should be removed.After this step,the calculation is simplified and the results are more accurate.11mintmOONsii (1)Where m is the row of binary image,and t1 is thickness factor of umbrella plate.After all
37、n groups of digital sequences are processed,the image is converted to many unconnected regions,and the binary insulators umbrella plates and other residual targets are extracted effectively.2.3.2 THE SECOND BINARY SHAPE PRIOR KNOWLEDGE The thickness of umbrella plate on the edge is less than that of
38、 the center,and it increases gradually from the edge to the center.Determine each unconnected region respectively by the second binary shape prior knowledge and make sure whether the region satisfies the condition of the second binary shape prior knowledge.If the region satisfies the conditions,keep
39、 it;otherwise,eliminate it.Algorithm design is as follows:a)Number all the regions as 1,2,3,u;b)Count the number of 1 for each column recorded as Gi in region j,j1,u,save it as a vector G=G1,G2,Gi,Gp where p is the number of nonzero columns,compute the differences between two elements of G and get a
40、 vector H whose length is p-1 shown in equation(2).iiiGGH1 (2)If all the elements in H have the same sign(positive or negative),that means the thickness of the region increases(positive)or decreases(negative)gradually and the region can be judged and saved as insulator preliminarily.Also the region
41、containing of 0 in H would be kept in case of missing parts of insulator.c)Further remove impurities according to the position of first occurrence of 1.Search and record J=J1,J2,Ji,Jp as the position of the first occurrence of 1,and compute the differences between two elements of J and get a vector
42、K whose length is p-1 shown in equation(3).iiiJJK1 (3)The region j can be defined and saved as umbrella plate when all the elements in K have opposite signs with the elements in H.d)Search all the regions and remove the region which is not fit for the second binary shape prior knowledge based on the
43、 criterion of step b)and c).Then most of impurities would be removed,and insulator is saved integrallty except tiny parts.2.3.3 THE THIRD BINARY SHAPE PRIOR KNOWLEDGE The values of spacing between umbrella plates are small and approximate.Extract each column of the image respectively and obtain n gr
44、oups of 0-1 sequences l1,l2,lj,ln.Search the pixel points changed from 0 to 1 in sequence lj and mark their positions,and count the numbers of consecutive 0s before marked points as R1,R2,Ri,Rw composed of R where w is the number of R.The consecutive 0s can be defined as spacing of umbrella plates a
45、nd set to 1s if R meets the condition of equation(4)where Vmin is the least possible number of the interval,otherwise the consecutive 0s are not belong to insulator.21mintmRRVwii (4)Where m is the row of binary image,and t2 is the spacing factor of umbrella plate.After processed by the third binary
46、shape prior knowledge,the aerial image has a connected region at the location of insulator.4)The removal of small regions There is a large connected region at insulators location and there are small impurities at other locations.Set an appropriate threshold and delete the regions whose areas are les
47、s than the threshold,and then obtain a binary image of insulator.2.4 THE LOCALIZATION OF INSULATORS Rotate the binary insulator image with the angle at positive direction,and obtain the location of insulator in original aerial image.Mark insulator by minimum circumscribed rectangular box,and the ins
48、ulator can be localized in the aerial image.After all the possible orientation angles are traversed,all the insulators with different orientation angles can be localized in aerial image.3 RESULTS AND DISCUSSIONS 3.1 DETAILED ILLUSTRATION OF THE METHOD Take aerial images as examples,the sequential st
49、eps of the method are explained below:Figure 3a is aerial image(Im 1,size:353*640)with one insulator in complex background,and 3b is its binary image after preprocessing.It can be seen that the edges of binary image become smooth and clear.Sharpen the concave-convex and irregular curve of binary ima
50、ge into many connected contours composed by straight line segments shown in Figure 4a,and link edge pixels 3424Z.Zhao et al.:Localization of Multiple Insulators by Orientation Angle Detection together into a list of sequential edge points for each contour.For each contour,compute the candidate point