Explore Flipsnack. Transform boring PDFs into engaging digital flipbooks. Share, engage, and track performance in the same platform.
From magazines to catalogs or private internal documents, you can make any page-flip publication look stunning with Flipsnack.
Check out examples from our customers. Digital magazines, zines, ebooks, booklets, flyers & more.
Pre-made templates to create stunning publications in minutes
Here are eight reasons why you should consider choosing interactive, digital flipbooks instead of boring and static PDFs. Check them out!
MELGANIANDBRUZZONE:CLASSIFICATIONOFHYPERSPECTRALREMOTESENSINGIMAGESWITHSVMs1779 classi edbyusingtheavailabletrainingsamples.Then,theclas- si edsamples,togetherwiththetrainingones,areexploitedit- erativelytoupdatetheclassstatisticsand,accordingly,there- sultsoftheclassi cationuptoconvergence[4],[5].Theprocess ofintegrationbetweenthesetwotypologiesofsamples(i.e., thetrainingandthesemilabeledsamples)iscarriedoutbythe expectation – maximization(EM)algorithm,whichrepresentsa generalandpowerfulsolutiontotheproblemofMLestimation ofstatisticsinthepresenceofincompletedata[6],[7].Themain advantageofthisapproachisthatit tsthetrueclassdistribu- tionsbetter,sincealargerportionoftheimage(availablewithno extracost)contributestotheestimationprocess.Themainprob- lemsrelatedtothissecondapproacharetwo:1)itisdemanding fromthecomputationalpointofviewand2)itrequiresthatthe initialclassmodelestimatedfromthetrainingsamplesshould matchwellenoughtheunlabeledsamplesinordertoavoiddi- vergenceoftheestimationprocessand,accordingly,toimprove theaccuracyofthemodelparameterestimation. Inordertoovercometheproblemofthecurseofdimension- ality,thethirdapproachproposestoreducethedimensionality ofthefeaturespacebymeansoffeatureselectionorextraction techniques.Feature-selectiontechniquesperformareductionof spectralchannelsbyselectingarepresentativesubsetoforiginal features.Thiscanbedonefollowing:1)aselectioncriterionand 2)asearchstrategy.Theformeraimsatassessingthediscrim- inationcapabilitiesofagivensubsetoffeaturesaccordingto statisticaldistancemeasuresamongclasses(e.g.,Bhattacharyya distance,Jeffries – Matusitadistance,andthetransformeddiver- gencemeasure[8],[9]).Thelatterplaysacrucialroleinhyper- dimensionalspaces,sinceitde nestheoptimizationapproach necessarytoidentifythebest(oragood)subsetoffeaturesac- cordingtotheusedselectioncriterion.Sincetheidenti cationof theoptimalsolutioniscomputationallyunfeasible,techniques thatleadtosuboptimalsolutionsarenormallyused.Among thesearchstrategiesproposedintheliterature,itisworthmen- tioningthebasicsequentialforwardselection(SFS)[10],the moreeffectivesequentialforward oatingselection[11],and thesteepestascent(SA)techniques[12].Thefeature-extraction approachaddressestheproblemoffeaturereductionbytrans- formingtheoriginalfeaturespaceintoaspaceofalowerdi- mensionality,whichcontainsmostoftheoriginalinformation. Inthiscontext,thedecisionboundaryfeatureextraction(DBFE) method[13]hasprovedtobeaveryeffectivemethod,capable ofprovidingaminimumnumberoftransformedfeaturesthat achievegoodclassi cationaccuracy.However,thisfeature-ex- tractiontechniquesuffersfromhighcomputationalcomplexity, whichmakesitoftenunpractical.Thisproblemcanbeover- comebycouplingwiththeprojectionpursuit(PP)algorithm [14],whichplaystheroleofapreprocessortotheDBFEby applyingapreliminarylimitedreductionofthefeaturespace with(hopefully)analmostnegligibleinformationloss.Anal- ternativefeature-extractionmethod,whoseclass-speci cnature makesitparticularlyattractive,wasproposedbyKumar etal. [15].Itisbasedonacombinationofsubsetsof(highlycorre- lated)adjacentbandsintofewerfeaturesbymeansoftop-down andbottom-upalgorithms.Ingeneral,itisevidentthatevenif feature-reductiontechniquestakecareoflimitingthelossofin- formation,thislossisoftenunavoidableandmayhaveanega- tiveimpactonclassi cationaccuracy. Finally,theapproachinheritedfromspectroscopicmethods inanalyticalchemistrytodealwithhyperspectraldataisworth mentioning.Theideabehindthisapproachisthatoflooking attheresponsefromeachpixelinthehyperspectralimageas a one -dimensionalspectralsignal(signature).Eachinformation classismodeledbysomedescriptorsoftheshapeofitsspectra [16],[17].Themeritofthisapproachisthatitsigni cantlysim- pli estheformulationofthehyperspectraldataclassi cation problem.However,additionalworkisrequiredto ndoutappro- priateshapedescriptorscapableofcapturingthespectralshape variabilityrelatedtoeachinformationclassaccurately. Othermethodsalsoexistthatarenotincludedinthegroup ofthefourmainapproachesdiscussedabove.Inparticular,itis interestingtomentionthemethodbasedonthecombinationof differentclassi ers[18]andthatbasedoncluster-spacerepre- sentation[19]. Recently,particularattentionhasbeendedicatedtosupport vectormachines(SVMs)fortheclassi cationofmultispectral remotesensingimages[20] – [22].SVMshaveoftenbeenfound toprovidehigherclassi cationaccuraciesthanotherwidely usedpatternrecognitiontechniques,suchasthemaximum likelihoodandthemultilayerperceptronneuralnetworkclassi- ers.Furthermore,SVMsappeartobeespeciallyadvantageous inthepresenceofheterogeneousclassesforwhichonlyfew trainingsamplesareavailable.Inthecontextofhyperspectral imageclassi cation,somepioneeringexperimentalinvestiga- tionspreliminarilypointedouttheeffectivenessofSVMsto analyzehyperspectraldatadirectlyinthehyperdimensional featurespace,withouttheneedofanyfeature-reductionpro- cedure[23] – [26].Inparticular,in[24],theauthorsfound thatasigni cantimprovementofclassi cationaccuracycan beobtainedbySVMswithrespecttotheresultsachieved bythebasicminimal-distance-to-meansclassi erandthose reportedin[3].Inordertoshowitsrelativelylowsensitivity tothenumberoftrainingsamples,theaccuracyoftheSVM classi erwasestimatedonthebasisofdifferentproportions betweenthenumberoftrainingandtestsamples.Aswillbe explainedinthefollowingsection,thismainlydependsonthe factthatSVMsimplementaclassi cationstrategythatexploits amargin-based “ geometrical ” criterionratherthanapurely “ statistical ” criterion.Inotherwords,SVMsdonotrequire anestimationofthestatisticaldistributionsofclassestocarry outtheclassi cationtask,buttheyde netheclassi cation modelbyexploitingtheconceptofmarginmaximization.The growinginterestinSVMs[27] – [30]iscon rmedbytheirsuc- cessfulimplementationinnumerousotherpatternrecognition applicationssuchasbiomedicalimaging[31],imagecompres- sion[32],andthree-dimensionalobjectrecognition[33].Such aninterestisjusti edbythreemaingeneralreasons:1)their intrinsiceffectivenesswithrespecttotraditionalclassi ers, whichresultsinhighclassi cationaccuraciesandverygood generalizationcapabilities;2)thelimitedeffortrequiredfor architecturedesign(i.e.,theyinvolvefewcontrolparameters); and3)thepossibilityofsolvingthelearningproblemaccording tolinearlyconstrainedquadraticprogramming(QP)methods (whichhavebeenstudiedintenselyinthescienti cliterature). However,amajordrawbackofSVMsisthat,fromatheoretical pointofview,theywereoriginallydevelopedtosolvebinary
The cookies we use on Flipsnack's website help us provide a better experience for you, track how our website is used, and show you relevant advertising. If you want to learn more about the cookies we're using, make sure to check our Cookie policy
We use essential cookies to make our site work for you. These allow you to navigate and operate on our website.
We use performance cookies to understand how you interact with our site.They help us understand what content is most valued and how visitors move around the site, helping us improve the service we offer you.
We use marketing cookies to deliver ads we think you'll like.They allow us to measure the effectiveness of the ads that are relevant for you.