Automatic Satellite's Streak Detection In Astronomical Images Based On .
Journal of the Earth and Space Physics, Vol. 46, No. 4, Winter 2021, P. 49-64 (Research)DOI: 10.22059/jesphys.2020.281849.1007120Automatic Satellite’s Streak Detection in Astronomical Images Based on IntelligentMethodsFarzaneh, S.1*, Sharifi, M. A.2 and Kosary, M.31. Assistant Professor, Department of Surveying and Geomatics Engineering, Faculty of Engineering, University ofTehran, Iran2. Associate Professor, Department of Surveying and Geomatics Engineering, Faculty of Engineering, University ofTehran, Iran3. Ph.D. Student, Department of Surveying and Geomatics Engineering, Faculty of Engineering, University of Tehran,Iran(Received: 25 May 2019, Accepted: 29 Sep 2020)AbstractThe orbit determination in one sentence is the application of a variety of techniques for estimatingthe orbits of objects such as the moon, planets and spacecraft. In dynamic astronomy, the orbitdetermination is the process of determining orbital parameters with observations. Considering thevisibility of the satellite motion trace and the fundamental need to determine and modify satellites’orbital parameters as well as identify special satellites, determining the positional parameters of thesatellite is also one of the modern and important applications of vision-based astronomicalsystems. In the modern vision-based astronomical systems, data collection is done using a chargecoupled device (CCD) array. In this paper, a new method is presented for satellite streak detectionthrough an optical imaging system. This automatic and efficient method, which has the ability ofreal-time data analysis, is based on the sidereal image using CCDs. The images captured by thismethod have a large amount of information about stars, galaxy, and satellites’ streaks. In thispaper, an automatic method is presented for streak detection. The purpose of this research is to findan optimal method for satellite streak detection and different methods in clustering such ask means, particle swarm optimization (PSO), genetic algorithm (GA), and Gaussian mixturemodel (GMM). Finally, some assessment criteria were compared and concluded that GA is anoptimal algorithm in satellite streak detection.Keywords: Satellite tracking, Satellite streak detection, MSAC, Clustering, Swarm intelligence.1. IntroductionIn recent years, a limited number of countrieshave achieved some advancements in theaerospace industry and the ability ofbuilding, launching, and infusing satellites inlow orbits. In order to complete the entirecycle of the space industry, the satellitenavigation, and control, which has beenneglected since the beginning of themovement of space science in the country,has to be considered specifically. The orbitdetermination can be expressed as theapplication of a variety of techniques forestimating the orbits of objects such as themoon, planets, and spacecraft. Today,satellite orbit determination is known asprimary part of space surveillance in designand control of them after launch. Therefore,the ability of tracking and satellite orbitdetermination is among the most importantissues in the space program of each country.From ancient times, orbit determination hasbeen a challenge for space scientists. Due to*Corresponding author:increasing space missions and the number ofsatellites, it is necessary to establish newmethods for accurate detection of orbits andidentify spy satellites. In particular, orbitdetermination of planet of the solar system isan adjustment of noisy orbital observationthat consists of random and systematic errorfor force models and estimation of modelparameters by observations, such that toachieve a mathematical model that illustratesthe path of the celestial object in the pathbefore and after the observation time. Tosimplify, this process is divided into twoparts. First, the initial orbit is estimated andthen corrections are made to the determinedorbit. The purpose of initial orbitdetermination of the object moving aroundthe earth is to calculate object orbitalparameters by few observations; furthermore,initial orbit determination is used fordetecting a missing object in space. Todetermine the precise orbit, it is necessary tofarzaneh@ut.ac.ir
50Journal of the Earth and Space Physics, Vol. 46, No. 4, Winter 2021determine the initial orbit with goodaccuracy, which indicates the importance ofthe initial orbit determination (Vallado andAgapov 2010; Swanzy, 2007; Farnocchia etal., 2010).In general, tracking and orbit determinationprocedure can be divided into two sections:observation and calculation. The first isextraction of position and velocity of thesatellite and in part of the calculation,numerous methods and algorithms abouttracking and satellite orbit determination areappliedtothem.Satellitealtitudedetermination and positioning requiressufficient information on physics of themotion and other controlling parameters.Since the physics of earth and atmosphere isnot adequately known, orbit determination isnot possible with the available model.Therefore, for a highly accurate orbitdetermination of civilian satellite, it isnecessary to directly observe the satellite.Different types of observations are used tomake an initial orbit determination. Theseobservations can be collected by groundstations that contain angular angles,elevations, distance and distance range.These observations are made by the radar andthe telescope because the collection ofobservations without an instrument and thenaked eye does not have enough precisionand sensitivity for the determination of thespace object orbit. However, because thedistante observation is expensive andsometimes impossible, angular observation isused. Thus, the optical tracking system ismore accurate and simpler and requires lowcost equipment. Accordingly, optical l condition (Lee, 2003; Lee etal., 2004).Fundamental of orbit determination usingoptical system is astronomical imaging usinga CCD. Images captured with this methodcontain large amount of information aboutstars, galaxies, and satellite streak(Schildknecht1994,2007).Orbitdetermination method based on this methodused stars’ position for attitude determinationof satellite, so coordinates of satellite andstars should be detected in images. Since asatellite appears as a streak in the capturedimage, the model of the streak satellite mustbe extracted accurately, because themisdiagnosis of the beginning and end pointsof the streak directly affect the accuracy ofthe determined orbit. Automatic method isimperative for streak detection becausesatellite recognition manually is tedious,time-consuming, and erroneous. Manualstreak detection was used in initial methodswhen technology was not available or it wasnot affordable. With the advancement oftechnology, the dependency on manualmethods has drastically decreased (Hejduk etal., 2004).In satellite detection, streak can sever as aline such that the efficient line detectionmethod affects the diagnosis of beginningand end points. There are some studiesregarding line detection in image processing.The first method in this regard was proposedby Hough in 1962 as a Hough Transform(HT) algorithm, which transfers an edgeimage in parameter space. Another type ofHT is Probabilistic Hough Transform thatdetermines the start and end points of lines.Lévesque performed the matched filteralgorithm for satellite streak detectiondetection. (Levesque, 2009; Levesque andButeau, 2007). Also, a number of algorithmshave been developed for line detection usingPSO, which performs faster than HT (Simms,2011; Kirchmaier et al., 2010).There is a need to develop algorithms andsoftware that can automatically detect andreport the presence of satellite streaks andstart and end points in the acquired images.The algorithms presented in this paper weredeveloped for this purpose. The mainobjective of this study is to develop a streakdetection algorithm based on the clustering.The basic idea of the method is to usenumerous clustering algorithms to improveaccuracy and speed and decrease run-time.The remainder of this paper is organized asfollows. Section 2 describes the proposedmethodology. Section 3 provides aquantitative and comparative experimentalvalidation of the proposed approach usingsimulated and real astronomical images.Results of streak detection on two datasetsare presented in Section 4, and conclusionsare provided in Section 5.2. MethodAn automatic streak detection for opticalimages is developed in this study. In addition
Automatic Satellite’sSStreeak Detection in Astronomicaal Images Baseed on to sttreaks of thetsatellite,, the obserrvedimagees contain noise,nstar, annd space debbris.Thereefore, it is imperativeitot illuminatee allnon-sstreak compponents froom the immagebeforre the streaak detectionn algorithmm isperfoormed on thee image. Thee overall proccessincluddes four stepps: 1) noise reduction inn theopticaal image, 2) extraction of star centerss, 3)star removal, and 4) clustering andsegmmentation. Figgure 1 illusttrates the gloobalproceessing. The basic approoach consistss ofusingg PSO to deteect the streakk.Figuree 1. General procedure forr satellite’s sttreakdetectionn.2-1. NNoise reducttionIn thhe modern vision-baseed astronommicalsystemms, data colllection is donedusing CCCDarrayy. During thee process of lightlcollisioon tothe surface of the CCD, reading andmeasuring the nuumber of phootoelectrons, andconveerting them to the digiital numberss tostore them as greey degree inn each pixel,, thesmalllest mistakes that resultt in the losss oraddinng of electronns on each pixelpcan leaad todistorrtion and nooise in the immage. Noise canbe ggenerated byy external factorsfsuchh astempeerature andd physical conditions, orinternnal factors associatedawith CCD. NNoisereducction is perfoformed as a primary stepp inproceessing. Thee noise reemoval proccessshoulld not onlyy eliminate or reduce thenoisee, but also avvoid blurringg the image andremoving or rellocating thee edges of theimagee (Gonzalezz, 2009). Too determine theprimaary orbit of thet satellite using an optticalmethood, the streak of the saatellite mustt beextracctedacccuratelybecause,bthe51misdiagnosis of the begiinning and ennd pointsof the streakk directly affffects the acccuracy ofthe determinned orbit. TTherefore, a methodshould be used that imposes thee lowestpossible effefects to the kkey compliccations ofthe astronommical imagees such as star andsatellite strreak. There are severral noisereduction methodsmsucch as averaage filter(Said et al., 2016), mmedian filterr (AriasCastro and Donoho, 22009), and Gaussianfilter (Buadees et al., 20005). Negativve pointsof these filteers are smooothing and coonvertingthe edge witth simultaneoous removal of noise,so they are not approprriate for optical orbitdeterminatioon.If the gray leevels of the iimage are coonsideredas temperatture, then hheat transferr processdecreases thhe gray levells over time and thushelps noise reduction.rInn order to preeserve theedges, the thhermal conduuctivity coeffficient isdetermined by the possition of thhese graylevels. This method prrovides consstancy ofboth edge shhapes and nooise removal. Thermalconductivityy coefficiennt depends on theposition of current andd surroundinng pixels.Therefore, thetimage is smoothedd in thedirection off edges. Hoowever, theyy do notchange in thhe perpendicuular directionn to themand thus thee position annd direction of edgesremain unchhanged (Weeeratunga andd Kamath2002, 2003)). In this studdy, it is atteempted toeliminate thee noise in thee optical imaage usingthe diffusioon equationn. Furthermmore, toidentify the accurate poosition of thhe edges,the gradiennt is calcuulated through theconvolution of the mmain image by theGaussian filtter.There is a widew range off methods to solve thediffusion equation.eInn this stuudy, thisequation waas solved bby an iteratiion-basednumerical methodm(Geerig et al., 1992).Generally, thhe more thee paces and iterationsin the equaation, the ssmoother thhe imagewould be. OneO parameeter must bee chosensuch that the image brightness doesdnotexceed the mainmrange. For this purrpose, thenoise must beb eliminateed from the imageibychoosing an appropriate number of iterations.In this research, the structural similaritysindex (SSI) is used to select the optimumnumber of iterations. TThis index considerscgray level changes. Also, it expressesecontrast innformation and structtures incomparison of images. TThus, SSI is based on
52Journal of the EarthEand Spacce Physics, Vool. 46, No. 4, WinterW2021a weightted combinattion of three criteria suchhas luminnance ( . ),contrast ( . ), anddstructuree ( . ), asa stated in Equation (1 )(Wang, 22017; Wang et al., 2004)). . .(1))2-2. Dettection and Removal off Stars frommthe ImageThe secoond step in thet proposedd algorithm issstar cennter extractioon. In ordeer to extracctthese oobjects, sccale invariant featureetransformm (SIFT) havve been usedd as an indexxto descrribe the locaal feature inn the imagee.These ffeatures aree invariant to uniformmscaling, orientation, illumination changess,and parrtially invarriant to loccal geodeticcdistortion (Lowe, 19999). In this method, theefeatures are extractted efficientlly and thenn,through a multi-stage filtering method,mstableepoints (ii.e., key points) are ideentified in ascaled sppace. In this algorithm, without theeneed to tthe distributiion function,, the positionnof the sstar center isi determineed with subbpixel pprecision auutomatically. Figure 2presents an overvview of staar detectionnprocedurres.Because star appeaared as a circlecin theeastronommical imagee, it is necessaryntooidentify pixels that are located in this circleeand deteermine their centers using the SIFTTalgorithmm. The probblem faced inn this regarddis that rradii of the circles are unknown;usoothey werre calculatedd repeatedly. This meanssthat for ddifferent raddii, the mean gray level oofpixels ouutside the cirrcle was calcculated and aradius wwith no brigght pixels ouutside it wassselected.Figure 22. Star recoggnition process using SIFTTalgorithm.2-33. Satellite Sttreak DetecttionAs noted prevviously, sincce satellite streakapppears as a line in the astroonomical image, itis immperative too detect start and end poiints oflinee precisely. Therefore, determination ofmatthematical modelmfor sattellite streak playsa keyk role in thistproceduure. In this study,sateellite streakk was deetected usinng aclustering algorrithm.Cluustering is one of tthe unsupervisedclasssification methodsmand a set of objeects isgroouped in somme classes aautomaticallyy. Thepurrpose of any clusteringg algorithm is toevoolve a dataset in such a wway that objeects inthe same class are based on the minnimumsquuared distannce criterioon and disstancebetwween diffeerent classses shouldd bemaxximum. Daata clusterinng is consiideredamong the most commonn techniquees forstattistical data analysis; fuurthermore, it canbe used in a widewrange oof issues suuch asmachine learning, pattern recognition, dataminning, imagee analysis, and other fields(Grrira et al., 2004).Ammong varioous clusterring algoriithms,hierrarchical annd density--based partiitionalclustering metthods are widely useed indiffferent topicss. These meethods, whicch areofteen based onn the cost function, emmployopttimization allgorithms foor clusteringg. Theopttimum classiifier is founnd by minimmizingthe cost function (Grira et alal., 2004).ms for dataa clustering haveMaany algorithmbeeen proposedd. In this rregard, the kmeansmetthod is one of the commmon and simplesmetthods (Jain, 2010). Recenently, heuristic andevoolutionary allgorithms haave been uttilizedfor data analysis and optimmization algoorithminsppired by PSO. As onne of the newestngroowing methoods, PSO iis a functioon ofcollective behaavior in artiificial intelliigencem, 2009). TheT theoreticcal foundatioons of(LimPSOO come fromm the behavvior of agentts thatinteeract locallyy with one another or theirenvvironments, suchsas somee insects (i.e., bee,ant, and termmite) or evven humanns. Inpoppulation, ageents have a simple struucturebutt their coollective beehavior cann becommplicated.Thee purpose ofo this studdy is to iddentifysateellite streak as a clusteriing problem usingPSOO algorithm and Gaussiaan mixture modelm(GMMM). In thee following, these algorrithmsare discussed.
Automatic Satellite’s Streak Detection in Astronomical Images Based on 2-3-1. The k-Means AlgorithmThe k-means is one of the simplestunsupervised clustering algorithms (Jain,2010). The main idea of this algorithm,which was presented by MacQueen in 1967,is to define k-center for each cluster throughthe following steps:- From N data, K data z1 , z2 ,., zk is chosenas cluster centers.- Assigns data xi , i 1, 2,., n to a specificcluster if Equation (2) is held:xi z j xi z p, p 1, 2, ., K , j p(2)- When all samples are assigned to clusters,the position of the K class centers isrecalculated. New class centers aredetermined by Equation (3):Zi 1ni x, i 1, 2, ., k, xj Cj(3)jKennedy, 1995).In PSO, agents in a particle have simplebehavior and follow achievement ofneighbors and themselves. The collectivebehavior that appears from this simplebehavior will cover optimal areas of a multidimensional search space. In PSO, a group ofparticle is used each having a strong potentialto solve the problem (Esmin et al., 2008).To update the particle’s position, it isnecessary to calculate the particle’s velocity.In PSO, in accordance with Equation (4), thevelocity consisted of three components:momentum, cognitive, and social (Van derMerwe and Engelbrecht, 2003).vij (t 1) vij (t ) c1r1 j (t )[ pbestij (t ) xij (t ) ] MomentumCognitivec2 r2 j (t )[ gbest (t ) xij (t ) ]Socialni illustrates the number of data in Ci class.- The second and third steps are repeateduntil class centers do not change.Although k-means have been developedduring the last decades, it cannot findoptimum solution accurately and has severalsignificant drawbacks such as:- The major disadvantage of the k-meansalgorithm is that the final solution dependson a number of clusters and their primarycenters; in addition, this algorithm issometimes stuck at suboptimal values.- There is no obvious procedure forcalculating the primary class centers.- If the number of data in one class is zero,there is no way to continue the method.2-3-2. Particle Swarm Optimization (PSO)AlgorithmPSO is a computational method thatoptimizes a problem iteratively based on asimulation of birds or fish collective behavior(Eberhart and Kennedy, 1995). PSO isoriginally attributed to Kennedy andEberhart. They first intended to develop akind of artificial intelligence using socialmodels that did not require individualabilities. Their primary simulation wasperformed in 1995 for the simulation of birdbehavior for finding seeds (Eberhart and53(4)In PSO, pbest and gbest play crucial roles inguiding the particle’s search. In the aboveequation, pbestij (t ) represents the bestposition that ith particle has experienced sincethe process has started, gbest (t ) representsthe best position of neighbors, and vij (t ) isthe previous velocity of the ith particle. xij (t )is particle’s position, c1 , c2 are accelerationconstant parameters that play the role ofweighting for cognitive and social parts, andr1 j (t ) and r2 j (t ) are random values between0 and 1, which are sampled from a uniformdistribution. Finally, the particle’s position ineach stage is updated according to Equation(5).xi (t 1) xi (t ) vi (t 1)(5)PSO algorithm is influenced by severalfactors such as data dimension, particlenumber, neighbor size, number of repetitions,and acceleration coefficients (Suganthan,1999).2-3-3. Genetic AlgorithmGenetic algorithm (GA) employs Darwinsnatural selection principles to find the
54Journal of the EarthEand Spacce Physics, Vool. 46, No. 4, WinterW2021optimal formula forr predicting or matchinggthe patttern. This algorithm is a searchhtechniquue to find thee optimal soluution and is atype of mmetaheuristiic algorithmss inspired byybiologicaal functionns such as mutationn,crossoveer, and selecttion (Coley, 1999).A GA involves severalsstepps. At firstt,dependinng on the datta and probleem, unknownnvariabless are specifieed. Then, these variablessare propperly encodeed and repreesented as achromossome. Basedd on the cosst function, asfitness ffunction is defineddfor chromosomescand, iniitially, the population is selecteddrandomly. Afterwardd, the value ofo the fitnesssfunctionn is calculateed for each chromosomece.Later, acccording to Figure 3, other steps areeperformeed sequenttially. Eacch step issdescribed in more details below (Coleyy,1999).ponn chromosommes resultingg fromis performeda crossovercoperator and wwith changinng thebitss of these chromosomescs make a wayw toenter new informmation.- AtA this stepp, the fitnesss value off newchrromosomes is calculatted in order toevaaluate the dauughter chrommosomes.- NewN populatioon is selectedd to enter the nextstepp of the algorithmabyy comparing thechrromosomes’ fitness valuees.- AllA the new populations are evaluatted. Ifthe algorithm is terminaated, it wiill befiniished; otherwwise, the currrent populattion isuseed as the innitial populaation for thee nextstepp.Thee terminationn condition oof the GA canc bedetermined by a problem (SScrucca, 20133).2-33-4. GMM AlgorithmAThee GMM is oneo of the nnovel methodds fordata analysis annd clusteringg. The satisfafactoryresuults with hiigh speed mmake this methodmverry useful. This methood is the moststattistically maature methood for clussteringbeccause each type of disstribution caan beappproximated by a sufficcient numbber ofGauussian functtion (Zivkovvic, 2004). In aGMMM, it is asssumed that eeach x is creatediinddependently byfolllowing densiity:amiixturewithhM p ( x l ) pibi ( x )the(6)i 1pi is a weightinng of model ith (by assummption0 pi 1forp pM1Figure 3. General oveerview of Gennetic algorithmmprocedure during sateellite’s streakkdetection.- In thhis step, a sufficient number oofchromossomes basedd on their fiitness valuessare seleccted and aree later usedd in the nexxtsteps.- The crrossover operrator with Pc probabilityyis applieed to parentt chromosommes and newwchromossomes are generated.- The mmutation operrator with Pm probabilityy 1)eaachandandk bi ( x )arred-dimmensional GaussianGdisttribution funnction witth average i and covariiance matrices i ,resppectively. bi ( x ) 1(2 )D/ 2 i1/ 2 (7) 1 exxp ( x i ) i 1 ( x i ) 2 Thee clustering processpthereeby is transfoormedsucch that to esstimate the pparameters ofo theGMMM. Expectaation-maximmization algoorithmis useduin this studysto estimmate the unkknownparrameters of thhe GMM (Ziivkovic, 2004).Thee output of clusteringcis a binary imaage inwhich only sattellite streakk pixels are found
Automatic Satellite’sSStreeak Detection in Astronomicaal Images Baseed on brighht.Somettimes,immagesinclludesomee satellite streaks thhat necessiitateseparrating eachh streak. Therefore, thepropoosed algoriithm involves connecctedcompponent labeliing process tot detect eachh ofthem (Suzuki et al., 2003). Since theree aresomee outliers inn streak cllass, RANSSACalgorrithm is exxecuted for line detecction(Fischhler and Bolles,B19881). With thisalgorrithm, the fittted line beccomes closeer tothe sttable positionn by some iteration and, as aresultt, start and end pointss of streak areidentiified precisely.3. Exxperimental resultThe theory of noisenreduction, star ceenterextraction, and streaak detectioon wasexplained inn Section 2. In this secction, theproposed meethod is impplemented onn the twoimages thatt have beenn used by CCDsCofiXON Ultrra 888 forr experimeents andanalyses. TheTdevice was equippped withback-illuminnated technoology and ann electronmagnifying setup to record thee singlephotons, thaat arrive at the CCD’s surface.Table 1 pressents the keyy specificatioons of themachine. As can be obbserved, the machinehas 1024 1024 square pixels withh a pixelsize of 13 μm.μ The highh reading rate of thissensor proviides an apprropriate capaability forconsecutive imaging of celestial boddies. Thestudied imagges are demoonstrated in FigureF4.Table 1. Keyy specification ofo iXON Ultra 8888(aa)55Active pixxel (H x V)1024 1024Pixel size (WW x H; μm )13 13 µmImage arrea (mm)13.3 13.3Actiive Area Pixell Well Depth (e( )e -80Max Readoutt Rate (MHz)30MHZFrame raates (fps)26(fufull frame)-96990Read nooise (e-) 1 with EM gainnQE MaxM 90%(b)(Figuree 4. sample imaages used in this study using Ixxon Ultra 888 CCDCwith differrent Exposure ttime a) 0.5 sec anda b)1sec.
56Journal of the EarthEand Spacce Physics, Vool. 46, No. 4, WinterW20213-1. Noise ReductionThe imaage noise redduction was implementeddon the iimages usingg the diffusiion equationn.As noteed previoussly, this methodmis afunctionn of the nuumber of iteration anddthis parrameter is unknown. In order toofind an optimal nuumber of iterations, theeSSI indeex was utilizzed. The rangge of 0 to 100is considdered as iteeration and thet expecteddvalue iss given byy the maximmum indexx.ult ofThee variation of SSI inddex and resunoise reductionn are presennted in Figuures 5andd 6.2. Star Extraaction3-2Staar extraction algorithm is applied tot thedennoised imagge and the SIFT algoorithmparrameters for two imagess presented ini thispapper (Table 2). Also, the center of sttars ispresented in Figgure 7.FFigure5. Variaation of structurral similarity inddex in range off 10 iteration.Figure 6. Result of removing noisse in sample immages using difffusion equationn.Table 2. The SIFT algorrithm parameteers for star extraaction.SIFFT sSigmaEdge Ratio2.11044451.55
Automatic Satellite’sSStreeak Detection in Astronomicaal Images Baseed on 57Figure 7. DisplayDof star’ss center pixel cooordinate using SIFT algorithmm.Afterr extraction star’s center, pixels hhavebeen occupied byb star are removed ffromimagees using prooposed methood, and Figuure 8is a reesult of this step.s3-3. Satellite Streak Deetection Ussingnce AlgorithhmsSwarrm IntelligenIn thiis study, fouur clustering algorithms ((i.e.,PSO, GA, GMMM, and k-means)kwwereimpleemented on images to detect sateellitestreakk cluster. GA and PSOPalgorithhmsrequire some initial parameeters, thus iit isnecesssary to callculate optimmal parameetersbeforre the clusterring. The nummber of iteraationand initial populaation play immportant rolees inthe PPSO methood and direectly affect thesolutiion. To determinedthhe number ofiteration for PSO and GGA, the probblem wasrun for 1 to 50 iterationss in one imaage. Also,the Sum off Squared EError (MSE) betweenbinary imagge after clusttering and thhe imagewith only saatellite streakk, each iteraation wasplotted in FigureF9. TThe plot shows thatadding moree iteration ddoes not impprove theresults conssiderably. TTherefore, 5 and 4iterations werewselecteed as the optimaliteration paarameter foor GA annd PSO,respectively. This proceddure is repeaated for anumber of populations,pbut the diffference isthat the prroblem was run for 1 to 100populations and the MSEE for each numbernofthe sample wasw presenteed in Figure 10. As aresult, 49 annd 7 samplees opt as thee optimalnumber of population.Figure 8. RemovingRstar ppixels from imaages using propposed algorithmm.
58Journal of the EarthEand Spacce Physics, Vool. 46, No. 4, WinterW2021Figure 9. MSE for differrent iteration ussing PSO and Genetic.GM for differeent population usinguPSO and Genetic.Figure 10. MSEns algorithmmAs stateed previoussly, k meansometimmes convergees to a local minimum.mInnorder too solve thiss problem, combinationnmethod bbetween k mmeans and PSSO algorithmmis proposed. In this procedure,pcenterscof theeng PSO, GAA,initial cllusters are caalculated usinand GGMM. Afteerward, cllustering issperrformed usinng the k mmeans algorithm.Eveentually, the satellite streak moddel isdetermined usiing RANSAAC methods. Theouttputs of seveen algorithmms (i.e., PSOO, GA,k mmeans,GMM,PSO K mmeans,GAA K means, and GMMM K means) are11to 17.illuustratedinnFiguresFigure 11. Detected satelllite’s streak using k means alggorithm.
Automatic Satellite’sSStreeak Detection in Astronomicaal Images Baseed on Figure 12. Detectedd satellite’s streeak using PSO algorithm.aFiguree 13. Detected ssatellite’s streakk using Geneticc algorithm.Figurre 14. Detected satellite’s streaak using GMM algorithm.Figure 15. DetectedDsatelliite’s streak usinng PSO and k mmeans algorithmm.59
60Journal of the EarthEand Spacce Physics, Vool. 46, No. 4, WinterW2021Fiigure 16. Detected satellite’s sstreak using Gennetic and k means algorithm.FFigure17. Deteccted satellite’s sstreak using GMMM and k meaans algorithm.ussion4. DiscuIt is oobvious thaat all algoorithms canndetect ssatellite streeak and noo significanntdifferencce is seenn visually. In ordeerto comppare methodds, it is alsso necessaryyto evaaluate quanntitative reesults. Theeapproachh used to coompare differrent methodssis to aassess somee precision parameterssusinng the conffusion matriix and execcutiontimme. In this study, the confusion matrixmis calculatedcbeetween the ooutputs illusstratedin the previous section and the binarybimaage in whhich satellitte streak pixelsare bright. The evaaluation criteriapresented in Taables 3 and 4 are describbed asfolllows.Table 3. Evaluation critteria with differrent methods inn the sample image (a) in Figurre (4).PSOgeneticEMGMMEKK-meansPSOO kmeansGenetic kmeaansEMGGMM kmeansMisclassificationMn0.00106.69 e -40.00159.0086 e -40.00130.00149.08 e rrformance ime3.5633.61611.6967.7805.8416.56012.352
Automatic Satellite’s Streak Detection in Astronomical Images Based on 61Table 4. Evaluation criteria with different methods in the sample ima
tal numbers each pixel, in the loss ixel can lea age. Noise actors such conditions, ith CCD. N primary step moval proc or reduce the image edges of determine using an opt tellite must ecause, eak Detection i ved ris. all age is cess the , 3) and bal of treak ical CD n to and and to , the or d to can as or Noise in cess the and the the ical be .
The Maddox Rod Test. Maddox Rod (vertical streak with horizontal rods). Patient estimates horizontal separation between light spot and vertical streak . Patient estimates vertical separation between light spot & horizontal streak. Patient fixates the right eye red horizontal streak &
streak of the season. He opened the season on a 14-game reached base streak and followed that with streaks of 15 and 14 games. He is now on a 14-game stretch. Allen posted a four-hit night versus Central Michigan for his 24th multi-hit game of the season. The 4-for-5 effort moved his hitting streak to a season-best eight games.
From The Definitive Guide to Position Sizing Strategies Expected Losing Streaks in 100 Trades (or Losing Streaks as a Function of Winning Percentage of Trades) System Winning Trade Percentage . 100% Probability for a Losing Streak This Long Average Losing Streak Length . 10% Probability for a Losing Streak
yellow flecks to extensive areas of dark brown dry rot and become more severe as the plant ages. Cassava brown streak disease symptoms in roots of the susceptible variety, TME 204 (Namulonge, Uganda). Root constrictions and malformation are a common feature of severe CBSD. Cassava brown streak disease leaf symptoms in TME 14 (Namulonge, Uganda).
Washington State also has a senior quarterback in Jeff Tuel. BYU wide receiver cody hoffman has a reception streak of 19 games and counting, the 21st longest streak in the nation. Washington State wide receiver Marques Wilson has caught a pass in 24 straight, tied for the ninth-longest streak.
the satellite output power to the maximum level, additional noise is introduced into the link from satellite to hub. Therefore, an accurate calculation of the SNR for the entire RTN link must consider: 1. the SNR of the terminal-to-satellite link 2. the SNR of the satellite-to-hub link When the output power of the satellite is at a maximum, SNR .
c. Satellite: Internet access provided through satellites orbiting the Earth. Satellite service requires a satellite Internet subscription from an Internet satellite service provider and a satellite dish. Carriers that provide satellite Internet service are DIRECTV, Dish Network, HughesNet, and Wildblue. d.
phere SSA-850 RV Satellite Antenna with Auto skew TWIN LNB is innovative and a technologically advanced satellite Positioner system. The antenna has a unique combination of state-of-the art components with the most sophisticated satellite acquisition and tracking programs to locate and lock onto the OPTUS C1/D3 Satellite in
