IE E K S IV Page o f IEIE SPC IEIE Transactions on Smart Processing & Computing HOME About IEIE SPC CONTACT u s SITEMAP View Articles Editorial Board Recent Hot Papers View Articles (online) ISSN 2287-5255 E-journal Special Issues Manuscript Submission Author Guide Editor and Reviewer Guide w TRANSACTIONS ON SMART PROCESSING COMPUTING m Volume Number June ,2 Sm art Signal Processing Overview of Bitstream Syntax and Parser Description Languages for Media Codecs s.Ja n g 103 K h an h L e Q u a :, T an N gu yen S y , T ĩìan h N g u y en T ill N hat, a n d H a L e T han h 1j H yu n gyu K im a n d ilu e e Burned Area Detection After Wildfire Using Landsat ETM +SLC-off Im ages Single Im age Enhancement Using Inter-channel Correlation Jin K im , S o o w o o n g Je o n g , Y on g H o K im , a n d S a n g k eu n L e e 130 The Vaguelette-Curvelet Decomposition for Im age Deblurring C h an g h u n C hơ, A g g e ío s K K a ts a g g e to s , a n d Jo o n k i P a ik HO Ju n m in S h i, Yi S u n , X idO Chen Z h an g , a n d Jiz h o n q X iao 148 Sm art W ireless Com m unications Random Access Channel with Retransmission Gain Resource Allocation based on Quantized Feedback for TDMA Wireless Mesh Networks L e i X u, Z h en -m in T an g, Y a-p in g Li, Y u -tvan g Y an g S h a o -h u a ta n , a n d T o n g -n w g i v 160 Sm art Com puting An Escrow -Free Tw o-party Identity-based Key Agreement Protocol without Using Pairings for Distinct PKGs T fio k o z a m F elix V alien t, E u n -Ju n Y oon , a n d H yu n su n g K im 68 R a je e v S in g h a n d T e e k P arv a ! S h a rm a 176 A Secure WLAN Authentication Scheme IE1E T r a n s a c t i o n s on S m a r t Proc es sin g & C o m p u t i n g A publication o f the Institute o f Electronics and Inform ation E ngineers (IE1E) Rm 907, K orea S cience and T ec h n o lo g y Blclg , 22, T eh e n -ro 7-fill, Ciaimnam -m i, S eoul, S E O U L 135-703 R ep o f K O R E A P h o n e + -2 -5 -0 5 -7 , Fax + 82-2-552-6093 http // W W W leiespc org http://ieckspc.ieekweb.org/vievv_acticles/acticles_d.asp?j=10 8/7 /2 117 IEF.K Transactions on Smart Processing 1111(1 Computing, vol 2, no 3, June 2013 w • m m m — “ ikkM M ttm w m m m m I LI mmmmm Burned Area Detection After Wildfire Using Landsat ETM+ SLC-off images Khanh Le Quoc, Tan Nguyen Sy, Thanh Nguyen Thi Nhat, and Ha Le Thanh Human - Machine Interaction Laboratory, University of Engineering and Technology, Vietnam National University / Hamoi, Vietnam {Iqkhanh, tanns_54, thanhntn, ltha}@vnu.edu.vn * Corresponding Author: Khanh Le Quoc Received July 14, 2012; Revised July 28, 2012; Accepted August 14, 2012; Published June 30, 2013 * R egular Paper A b stra c t: T h e increasing d e m a n d for m o nito rin g wildfires and their im pact o n the land su rface h av e p ro m p ted studies o f bu rn ed area extraction and analysis T o differentiate bu rn ed a n d u n b u rn e d area, the earlier m eth o d o f the M o d erate R e solu tio n Im ag in g S p ectro-radiom eter ( M O D I S ) B u rn ed A rea D etection A lg orith m w as pro p o sed to estim ate the ch an g e in land surface b ased o n the reflectance energy T he energy, w h o se w a vele ng ths are sensitive to burn ing , w as selec ted to calculate the chan ge p a m e te r Z SC0IV T his m eth o d w as ap plied using the M O D 1S im ages to p ro d u c e a M O D I S B urn e d A rea product T h e a p proac h w as to sim plify this algorithm to m ake it c o m p a tib le with the L a n d sat E T M + S L C - o f f im ages T o extract the refined version o f burned regions, po st-p ro cessin g w as carried out by ap plyin g a m edian filter, dilation m o rp h o lo g y algorithm , and fin ally a g ap filling m ethod T he exp erim en tal results s h o w e d that the d etailed bu rn e d areas e x tra c te d from the p rop osed m etho d exh ib ited m ore spatial details than th ose o f the M O D IS B u rn e d p ro d u cts in the large u s areas T h e results also revealed the dis tin uo us dis tribution o f b u rn e d regions in V ie tn a m forests K eyw ords- Burned area detection, Landsat ETM+ SLC-off, Wildfire, Land surface change Introduction T h e increasing recog nitio n o f burn ing biom ass as a w id e sp read and significant agent o f clim ate and en viron m ental ch ang e has led to an on go in g nee d for lo ng ­ term fire data at the regional, continental an d global scale In part, this d e m a n d has been met w ith a substantial b od y o f satellite-based active fire ob servations m ad e using a n u m b e r o f coarse- and m ed iu m -re so lu tio n sensors, such as the A lo n g -T c k S ca nnin g R ad io m eter (A T R S ) [ l ] , the V isible and Infrared S can ner (V1RS) [2] and the M o d e te R esolution Im ag in g S pec tro -ra dio m e ter ( M O D IS ) [3], W hile active fire products capture inform ation regardin g the location and time o f fires burn ing at the time o f the satellite o verpass, they not generally allo w an estim ation o f the reliable b u rned area [4], L arge-scale m aps o f bu rned areas, such as on a continental or global This research was supported by Project "Integration of WebGIS and multi-date and multi resolution satellite image processing technologies for forest fire monitoring in Vietnam" of Vietnam National University, Hanoi (VNƯII) under grant number QGT-D 12.25 'I ’ ■ IEEK Transactions on Smart Processing and Computing scale, are essential for a w ide ran g e o f a p p licatio n s, particularly for aeroso l em issio n estim ations T h i s need has p ro m p te d the d e v e lo p m e n t o f n u m e r o u s sate llite-b a se d m eth o d s to detect b u rn e d areas, the m ajo rity o f w h ich not exploit active fire inform ation T h e G L O B S C A R global burned a re a p ro d u c t [5] w a s p ro d u c e d for thie y ear 00 using tw o d ifferen t alg orith m s, textual a n d fixed threshold, ap plied to A T S R -2 and A A T S R im agery [6, 7] dev elo p e d a p re d ictiv e bi-directional reflectance m odel a p p roa ch to locate b u rn ed areas daily using 0 -m M O D S im agery In add itio n, there are a lg o rith m s that suppilemenl the standard r e m o te ly -se n s e d indicators for burn imapping w ith active fire m ap s [8] u se d tw o different vegietation indices d erived fro m 16-day M O D 1S na dir B R D F a d ju s te d reflectance c o m p o s ite s to id entify burn scars in central R ussia over a 12 -y e a r period [9] p ro p o s e d an a p p r o a c h to m ap various b u rn e d areas on an annual basis usiing the 0 -m M O D IS -d a y reflectance c o m p o s ite s withi 1-km MOD1S active fire m ask s T h ese a lg o rith m s have s e v e r a l characteristics that m a k e th em c o m p lic a te d to u s e First, the m in im u m d etec tab le size o f an active fire is up tio 1000 118 Qiioc et al.: Bunted Area Detection After Wildjire Using Landsal ETM+ SLC-off Images tim es sm aller than the m in im u m d etec tab le size o f a burned area [4], which can lead to c o n tam in atio n in selecting burned training pixels A n o th er c ause o f burned pixel contam ination is active fire false alarm s, i.e c o m m is sio n errors S econd, the selection o f un bu rned training pixels also is tam inated by a rang e o f fire sizes W hen the fire is loo small to detect, the a b sen ce o f fires d etected at a particular location does not gu aran tee that this location is unburncd T h e a p proac h o f [10] largely ov e rc o m e s these issues T his a lg o rith m w a s used to calculate the persistent c h a n g e s d e riv e d from 00 -m M O D IS surface reflectance on a daily basis T his algo rith m w as then used to g en erate regional density functions using active fire m ap s to d ete r m in e i f these chan ges are bu rned or u n b u rn e d in the d a ys nea rest to the bu rn ed dale A ltho ug h the alg orithm e x p lo ite d the inform ation o f the active fire m ap s m o re fully, it could not c o v e r the w ide range o f sizes o f b u rn e d areas Small burned areas w ere d etected but not e x tr a c te d in detail due to the low resolution o f the 0 -m M O D I S reflectance T his p ap er presents an a p p ro a ch to d etec t and extract the bu rn ed areas that use the h ig h -re so lu tio n im ages acquired by the sensor on b o ard L an d sat satellite T he Landsat series o f satellites provide a d ata sou rce for land surface m ap pin g an d m on ito rin g [11], T h e L andsat sensors include the L andsat T hem a tic M a p p e r (T M ) , Landsat E n h an ce d T hem atic M ap per Plus ( E T M + ) and L a n d sat 1-5 Multi spectral S cann ers (M SS) T he L an d sat se n s o r was launched in F ebruary 2013, a n d d a ta b e c a m e available from April 2013 B ccause the available data o f L and sat does not adapt to m etho ds that require lo n g -te rm data, the data o f L an dsat 1-7 w as still in w id e sp r e a d use, particularly the data obtained from the Landsat ETM+ sensor B efo re the launch o f L a n d s a t satellite, the new est L andsat E T M + w as still fu nctio ning , ev en th o u g h it has substantially ex ceed ed its plann ed d esig n life A ltho ug h the im ages collected by the L an dsat a n d satellites are available at no charge, the large a m o u n t o f data o v er a lo ng-term p erio d ob tained from L a n d sa t is a reliable resource b eca u se this m etho d m o nito rs the b u rn in g areas along the time series This a p p ro a ch exploits the active fire m ap s fro m Fire In fo rm ation for R esource M a n a g e m e n t S y ste m (F IR M S ) [12] to locate the active fire positions T h e app ro ach , w hich is a sim plified version o f the ea rlie r m e th o d o f the M OD1S B urned A rea D etection [13], calc u lated the c h an ges in the b urned areas verifie d by the F I R M S active fire maps T h e ch an g es w ere c alc ulated in both the spatial and tem po ral d o m a in lo gen erate a c h a n g e d e nsity function suitable for d iscrim in atin g the b u rn e d -re la te d and un bu rn ed-re la ted areas T his alg o rith m identifies the date o f burn in g , to the nearest days w ith in the individual Landsat im ages at a 30-111 spatial resolution T h e re m a in d e r o f this p ap er is s tru ctu red as follows Section su m m a riz e s the M O D I S B u rn e d A rea D etection A lg o rith m and Section d escrib es the a p p ro a c h as a sim plified version o f M - B A D A w ith a d d in g p ost p rocessin g Section p resents the e x p e rim e n ts an d results, and S ectio n reports the conclusion Burned Area Detection Method Using MODỈS Images [13] T h e M O D I S B urned Area D etection A lg o rith m (M B A D A ) has been d evelo ped to detect and m ap global b urned areas T h e algo rithm app lied the Bi-directional reflection m o del-based , chan ge-detection ap pro ach to m ap the 0 -m location and the ap p ro x im ate d a y o f fire T he detection w as b ased on the rapid chan ges in the daily M O D I S reflection data along the tim e series T his m ethod n eed s to refer to the data o f the previo us season s or years o f a given location to d eterm ine i f the spatial extent is bu rn ed or not T he detection m ethod, form ed by the bid irectio nal reflectance m o del-based ch an g e detection algorithm , is used in dependently to each geo -lo cated pixel ov er a long time series o f reflectance ob serv atio ns [14, 15] T h e reflectance sen sed w ithin a m o v in g te m poral w in d o w o f a fixed n u m b e r o f days (16 days ill this context) can m ak e the pre dic ted reflectance on a su bseq uen t day T he pred ic ted reflectance is then c o m p a re d with the o b serv ed reflectance by calculating the statistical m easure o f difference T his m e asurem ent helps m odel the directional d e p en d en ce o f the reflectance to prov ide a sem i-physicalb ased m eth o d lo predict the chan ge in reflectance from prev iou s slates The algo rith m w o rks on the M O D I S 500-rn spatial resolution im ages o f b an d (841 nm - 876 nm), b and (1 23 11111 - 12 nm) an d band ( 105 nm - l5 n m ) T h e n ear-in frared and longer w av elen g th reflectance bands arc used b eca u se they are g enerally insensitive to sm o ke aero sols em itted from veg eta tio n fires [16] R oy [6] presen ted that M O D 1S bands and pro vid e the highest bu rn ed 01‘ unbu rn ed area d iscrim ination but MOD1S band prov ides little T his observation helped build a set o f param eters and co nd itio ns for M - B A D A to detect the bu rn ed areas globally M - B A D A used a bi-directional reflectance m od el-b ased, ch ang e-detection m ethod to map the 0 -m location and the ap p ro x im a te day o f fire T he reflectance o f the land surface d etected by any rem ote se n so r v aried as a function o f the su n-surface-senso r ang les an d w as describ ed by the B i-directional R eflectance D istribution Function (B R D F ) [14, 15], T his study sho w ed that B R D F effects reduced the difference b etw e en the b u rned -related an d u n bu rne d-related land surface, both in the individual b and s and in spectral band indices T he de cre ased reflectance due to fire was less than the variation in the pre-fire reflectance cau sed by B R D F effects T he B R D F model c alculated the predicted reflectance and uncertainties for the v iew in g and illum ination angles o f a su b seq u e n t ob servation T he p a m e te r o f the B R D F m odel w as inverted against reflectance ob servations sen sed ill a tem poral w in d o w o f a 16-day duration This param eter, called Z scort, w as used as a norm alized m easure related to the prob ability o f a n e w ob serv ation belo ng in g to the current set T h e z score w as then calculated for M O D IS b a n d s and b ecause these b and s p ro v id e high sensitivity to b u rn in g areas A n e w ob servation w as co n sid e red a bu rn ed zhand cand ida te ^ z t h r e s h o ld 01 if the Z scnn ^h a n d > Z h r M i met w h ere the criteria: Z ,M u M is a IEEK Transactions D ll ! 19 S m all Processing mill Computing, vol 2, no 3, June 2013 fixed in dependent threshold O w in g to the characteristics o f the w av eleng th s in ban ds and 5, the burning area cau sed a decrease in the reflectance o f these ban d s but less ch an g e in the band reflectance, w h ereas the persistent ch an ges in cloud, shad ow , or soil m o isture w ou ld have a sim ilar effect on both bands To redu ce the im pacts o f these tem p ora l factors, the c o m p uta tio n o f the difference b etw ee n b a n d 2, and b and was repeated ind epend ently for each geo-located pixel along the time series This algo rithm allo w e d calculations o f both forw ard s and b ack w ard s in time to v erify the bu rned candidates F u rtherm ore, the algo rithm co uld increase the duratio n o f the B R D F inversion w in d o w , started in 16 days, and c o m p u te Z ivrc, for several subseq uen t days T his ensured the burned can did a te po int had been truly detected Proposed Method For Burned Area After Fire Using Landsat ETM+ SLCoff Images 3.1 Image Collection T he sen sor o f Landsat E T M + g enerates an o bserv ation for each location based on the W o rld w id e R eference System -2 (W R S -2 ) path /ro w system Each o bservation consists o f im ages corresp o n d in g to bands o f L and sat E T M + sensor sh ow n in T ab le I (band includes im ages in low an d high quality) with an inform ation m etad ata file T h e im ages for any location are free o f charge Landsat7 E T M + im ages provide high spatial resolution but low tem poral resolution T he tem poral resolution ol' im ages, w hich is the tim e required to revisit a position, is 16 days T he te m poral resolution o f data is in adeq ua te in the case o f rapid ch an g e detection for a p articular location such as daily changes In addition, cloud, sn o w or oth er w e a th e r co nd itio ns c o u ld im pact the pro c ess o f land surface m onitoring N ev ertheless, this tem p oral resolution m a k e s it possible to ob serv e the changes in lan dsca pe o v e r a lon g-term period, particularly in this co ntex t o f detecting the c ng es in bu rn ed area T he b u rned area after a wildfire re m a in s for w eek s or m o n th s so that it can m o n ito r the landscape D espite the low tem po ral resolution, the high value o f the L an d sa t E T M + data can be attributed in part to the relatively high spatial resolution Table Landsat ETM + w avelen gth co rrespo n ding spatial resolution bands (30 m for E T M + d e p ic t e d in Table I) T h e high spaatial resolution is im portant f o r d etecting the ch ang es in objeects on the land surface a n d separatin g them from unchanpged lands T herefore, the lack o f in form ation o f a tempooral resolution is c o m p e n s a te d fo r by the detailed spaatial resolution in the L andsat E T M + data T h e data obtainned from this sen so r can be used to detect and extract the bu rn ed areas after w ildfires O n I s' M ay 00 3, the Landsat E T M + Scan L J n e C orrecto r (S L C ) fa iled perm anently T he SSLC co m p en s a ted for the fo r w a r d m otion o f the satellite dul l ing scanning T he u ltim ate result o f SIX ' failure, referred ICO as SLC-off, is that so m e p arts o f an E T M + im age are not scanned T he u n -s c a n n e d data affected m o s t o f the im aage w ith the scan gaps v a r y i n g in size from one pixel lo) 14 pixels a lon g the east a n d w est edges o f the scene, as shoown in Fig Fig 1(a) sh o w s the co m p lete d Landsat E7TM + S L C - o ff im age, and Fig 1(b) presents the cropped im aage fro m Fig 1(a) nea r the M id d le East edge Fig 11(b) illustrates the black g a p s inside an im age that represcents the m issing da ta c a u se d by S L C failure T o m o n ito r the b u r n e d areas in the Earth, the IFire Inform ation for R e so u rc e M a n a g e m e n t System (F1RNV1S) (a) and Band W avelength Range (nm) Spatial Resolution (ni) 450 - 520 30 520 - 600 30 630 - 690 30 770 - 900 30 5 - 1750 30 10400 - 12500 30 2090-2350 30 520-900 15 (b) Fig C om pleted im ag e and cropped im age w iti bhack gaps 120 Quoc et ai: Burned Area Detection After Wildfire Using Laudsat F.TM+ SI,C-oJ) Images [I2J w hich integrates the rem o te s en s in g and G eog rap hic Infill mation System (G IS ) T e c h n o lo g y , delivers the MOD1S hotspots an d global fire locations T he data o f fire locations originated from the sta n d a rd M O D I S products: M O D 14 M Y D I Fire and T h e rm a l A n om alies, w h ich are then processed by Land A tm o s p h e r e near Real-tim e Capability (LANC1Z) for an E a rth O b se rva tion Sy stem (HOS) T h e s e standard M O D I S p ro du cts, w ho se spatial resolution is I km, only rep resen t the cen te r o f the active fire location T he ex am p les o f the active fire locations are depleted ill Fig for the regions o f V ie tn am and the ne igh b oring countries Land,sat E T M + im ages and fire in form ation achieved from F IR M S w e re co llected to c a lc u la te the chan ge in land surface ca u se d by wildfire T h e c h a n g e was then used as the input for the burned area d ete c tio n and extraction algorithm T he Landsat im ag es and active fire inform ation, w hich co ntains the occu rren ce time and location, w ere g ro u p e d as a triple L a n d sa t observation T he triple ob servation that w e re se le c te d along the time series c o n siste d tw o o b se rv a tio n s o b ta in ed before the o ccu rren ce time and one o b serv a tio n obtained after this time T h e se three ob serv atio ns are n a m e d “P re-J ”, "P re2 " and " P o st" , respectively T his triple ob servation is then calibrated by c ro p p rocess to focus on the fire position and reduce the c o m p u ta tio n time T h e c ro p process divides each im ag e in o bserva tions into n X 11 equal tiles For the w hereas the M O D I S b a n d p ro vided little discrim ination T ab ic lists the w av elen gths o f these bands T o apply the M O D I S burned area d iscrim ination m eth o d to the Landsat E T M + data, the c o rre sp on din g L an dsat bands, w hose w a v e len g th s that w ere the m o st similar to those o f M O D I S data, i.e M O D I S band relatively equivalent to Landsal b and 4, w ere sclccled T herefore, each ob servation o f Landsat included three im ages o f ban ds 4, and T he cro p p e d im ages from the three observations, called Landsat triple ob servation, w ere then the input for (lie pro cess describe d in Fig T he first step w a s to calculate the ch an g e alon g the tim e series o f the triple observation T h e next w as to detect the bu rn ed point Finally, the bu rn ed area extraction w as established test area, this stu dy e m p irically d e te r m in e d w h e th e r the value o f three w as app rop riate fo r n O nly the tiles that sisted o f active fire location s o b tain e d from F IR M S w ere used Fig presents all the steps to the pre-process data c alculated using the selected Landsat b and s to d iscrim inate the b urne d-re lated and u n bu rn ed related area as follows: 3.2 Change Measurement The pro p o s ed m eth o d defines a p aram eter that represents the ch ang e in each pixel o v er time b ased on the spatial intensity T his param eter, called z m., is a n o rm alize d m e a s u re m e n t o f the ch a n g e ill en ergy in a p articular pixel the d ifferent land co ver type, the Zj Z s 111, is a variable p a ram ete r can be = AP = (Ppog - P p „ - 2) MODIS bands (hat were sensitive and insensitive to biom ass b u rning , w ere used to d etec t the bu rn ed-related and n o n -b u rn e d -re lated change w ith in the scene, respectively An analysis o f the ability o f the M O D IS land surface re flecta n ce bands lo d is c rim in ate b etw een b u rn e d and u n b u rn e d area s s h o w e d that M O D I S bands 2, an d provide the highest b u rn e d - u n b u rn e d discrim ination, T herefore, re presen ting the characteristics o f the land surface even u nd er the sa m e env iro n m en tal condition O w in g to (lie different radiation, the reflection and scattering en erg y o f x V (1) V w here: Z) : Z scorc value o f w avelen gth X Ppn,_i : E n e rg y o f observatio n that recorded before the fire occurrence Fig Active fire location on FIRM S 121 IEF.K Transactions on Smart Processing (tiul Computing, vol 2, no 3, June 20/3 Find the triple observation that covers this location along time series: Pre- and ] Post-observation The active fire location recorded by FIRMS The triple observation found No -> Slop I Crop images in each observations Fig W o rkflo w of the prep ro cessin g step Landsat triple observation Burned area extraction Fig W o rk flo w o f the burned area extraction T ab le M apping w avelen gth M O DIS and Landsat sensors bands betw een MODIS (nm) Landsat (nm) Band 4: 770 - 900 Band 5: - 1250 Band 6: - 1652 Band 5: 5 - 1750 p The c o v er type before the fire occurrence Z n K w as co m p u ted for each and eve ry valid pixel, w h ereas that for the invalid pixels w ere not calculated V alid pixels w ere scann ed with the valid value, and the invalid pixels in the black gaps w ere not s c a n n e d by the / ) at the spatial co ordinates (/, j ) w as then calculated by: z* ° '') = a y > re p re sen ts th e energy p ix e ls b a se d on the relationship b e tw e e n the spatial resolution o f M O D I S and L an d sa t im ages C o m p a re d to the I- k m resolution o f the M O D I S im ag es, the - m resolation o f the Landsat E T M + im ages w a s m uc h sm aller, —33 fold T h e ratio s h o w s that o n e pixel po int o f the M O D IS i m a g e s c o rresp on ds to the 3 x 3 w in d o w in the L a n d s a t " E T M + im ages T h erefo re, the p re -d efin ed n u m b e r o f the A'indow size m is 3 x 3 pixels w ith a n u m b e r n cf valid n eig h b o u r pixels T h e s e n e ig h b o u r p ix e ls w e re u se d lo calculate the p a m e te r p (i, j ) as follows: p ( i j ) = J/ 1-71 (3) t-1 w h ere p t is the intensity o f the l'h pixel B ec iu se the ự p o stÍ Ì , D - P p r e - Í Ì j ) ) p u S that the valid p ix els located in a w in d o w o f m x n : E n erg y o f observation that reco rd e d after the A P (tJ) p ( i, j) , , respectively cente red at pixel (/,_/) T h e w in d o w size m is prtdefiined Band 7: 2090 - 2350 S L C failure T he in d ependent Z ( p aram ete r and p characteristics o f the land c o v e r type is the intensity average and n n eig h b o r pixels T he n n e ig h b o r pixels, are fire o ccu rrence p : P a m e te r representin g the en erg y o f the land „ w h e re p , sl(i, / ) , P ,rc- Ả i , j ) is the in ten sity o f the v a lid pixel (/', /■) in o bse rv atio n p Band 2: 841 - Band 7: - 5 the — l2) param eter, p (i, j ) , rep resents the land c o v e r type before (June el ill.: liiimcd Area Detection After Wildfire Using Landsat ETM+ SLC-o/J'images 122 (a) (b) Fig Fire in synth etic co lor im age and correspo n ding / Ii)n -dom ain im age (a) synth etic im age of the fire A shland in M o ntana, us, (b) im age in z ,, dom ain the o c c u rre n c e o f fire, this p a m e te r w as calculated for tile P r e -la n d P re -2 observations T h e s e two o bservations sh o w dial the y are highlv sensitive to b u rned-related im p acts, w h ic h a ss u m e s tlint there is a minority o f u n b u rn c d -re la te d changes T h erefore, the energy d istribution o f tw o observ ation s b efore the fire was eq u iv alen t so that />(/, /) was c o m p u te d as the average value of p n, /) and (/, j ) co rresponding o b s e rv a tio n P r e - l and P re -2 , respectively Fig 5(b) gives (in example of a fire in the domain of the original c o lor im a ge in Fig 5(a) T he bu rned area by the fire is b rig hte r than the n eig h b o rin g reg io n that can be observ ed c le a rly by h u m a n eye e x am ine d the value of z , a|1(J re c o m m e n d Z lhnMd = 0.4 As sh o w n fro m the im ages in the z , ^threshoU or Z > zthres holt P ỉ ĩ ? * ĩ \ i j ) - pfẴ"-đ2 3.3 Burned Point Detection (5) > PB pSSd - * ( i.j) - PB P %d - 7(i.D (6) T h e selec tion o f the three bands, and in the L a n d sat E T M f im ages presents the im portant points T h e b u rn e d an d u n b u rn e d areas are clear lo d iscrim inate in the b a n d 4, im ages, w hereas there is little discrim ination in tile b a n d im ag es alon g the time series o f before and after the fire T h is o bservation p ro vid es the criteria (5)-(7) for a g iven p oin t /.)(/, /') to be a b u rn e d point, w h ere / / ( / / ) is the in tensity o f the point (/, j) in the im a ge o f band V o f o b se rv a tio n V, an d z , is the z noise c aused by the s h a d o w or cloud T he experim ents c o m p u ted in b an d I o f an o bservation A p a rtic u la r point that satisfies both criteria w ou ld be labeled as a b u rn e d point in a b u rn ed area The criterion 4.1, w h ich is k n o w n as In ter-C h an g e C om putation, e x a m in e s the cliaim e in the land surface b ased on the z ( value Each pixel w ith a value is greater than a p redefined was c o n s id e re d a bu rned point candidate The bu rn e d point ca n d id a te s that satisfy the tw o last equations w ere v erifie d as the burned points T h ese two equations, called In tra -C h a n g e C o m p u tatio n , co m p a r e the ch an ge b e tw een the in n e r band s o f an o b s e rv a tio n to exclu de the P » S( U ) - pB P?e%-7 ( i D > r ? T ~ s( i j ) - r B r T ~ 7( i j ) (7) 3.4 Burned Area Extraction Tile L andsat im ages had an issue that is the ap p earan ce o f black gaps c a u se d by SLC-failurc P revious studies [17, 18] p ro po sed filling the black gaps T h e y w ere ap plied to spectral im ages o f L an dsat T hese algorithm s require the reference L an d sa t im ages that are unaffected by SLC-off T h e S L C - o ff im ag es has been p ro d u ced since 2003 T herefore, only the im ages w ithout S L C - o ff w ere ob tained before 2003 O ne study [17] also p ro po sed using the reference L an dsat S L C - o f f im ages to nil the gaps To acc o m p lish this, a w ide range o f S L C - o f f im ages needs to be obtained T h e large data, w h ich is free o f cloud and in the sam e w eath er conditions, w as difficult to achieve, p articularly for regions in V ietn am , w hich w ere used in the present experim ents M oreov er, the gap-filling process in I IX| w as in high c o m p lex ity and had a high co m putation time T h erefo re, in this study, gap filling w as perform ed using probability dis tribution in the b urn ed area binary IEEK Transactions on Smart Processing and Computing, vol 2, no 3, June 201 i (a) 123 (b) (c) Fig Exam ple o f burned area detection and extraction (a) im age in Z score dom ain, (b) Z m r -dom ain im ag e a fte r applying th resho ld cutting, (c) binary im age co rrespo n ding im age (b) (c) (d) Fig Exam ple of noise rem oval and burned area extraction (a) b in ary im age o f the burned area d etectio n , (b) binary im age after gap filling, (c) binary im age after noise rem oval by m edian filter, (d) binary im age o f b urned area extracted after dilation m o rp h olo gy alg orith m im ages instead o f the original spectral im ages H ere, the valid pixels are the pixels with the valid v alues in spectral im ages w h ereas the invalid pixels w ere in the gaps F o r each invalid pixel, a w in d o w c entered at this pixel w as formed to estim ate the value o f the pixel B ecau se the m a x im u m gap w idth is 14 pixels [18], a w in d o w size o f x pixels w a s used T h e value o f the invalid pixel w as estim ated as follows v _ fl >•0 rand < h / w (8) othemm w h e re n d Q is a function returning a un ifo rm d istributed n u m b e r in [0,1], b is the n u m b e r o f b urned p ix els in the w in d o w , a n d vv is the total n u m b e r o f valid p ix els in the w in w T h is estim ation function w a s d e riv ed u n der the ass u m p tio n that the invalid pixel shares the sa m e statistical properties o f b eing burned or u n bu rn ed with its neighbors ICEIC 2013, Jan 30 - Feb 2, 2013, Bali Indonesia fỉ —T z ỵ' s l / : Ỉ i -£—*— i — ẳ— ir Inpul LiQhlrtess Fig M a p p in g fu nction of e r e sp o n s e (dotted line), H E (da shed line), an d L ightness e n h a n c e m e n t fu nction (solid line) + a II XH Fig E n h a n c e d im a ges u n d er 12000 lux : (a ) input lin a g e , (b) IA C R M , (c) prop osed m e t h o d X = , and (d) Ả = 2} w w h ere is the d iag o n al m atrix in d ic a tin g a ratio b etw een the n u m b ers o f p o ssib le c o lo rs in the input lightness an d the tran sfo rm ed lig h tn ess, and a is a p aram eter to c o n tro l the deg ree o f lightness e n h an cem en t as sh o w n in Fig Ill the seco n d step , w e estim ate the am o u n t o f red u ced ch ro m a due to th e flare [3], T h e c o m p en sa te d chro m a c is o b tain ed by ad d in g the re d u c e d ch ro m a in C IE L A B : c = 2C K/ c,kf r a Propoted G\ Vi Ò to Ui to o\ z: o IJ L/Ì to Oi 'J I to 00 z D < Ọ c? o (O p to L/1 o OJ o ƠQ g •p( a>>? _) z f/Q •o o o o (— o p( fra rin •?s> to I -J to 'JI to 'J I 10 f/Q - — co^ H !—>* p —J z sc > 50' z CƠ o be PI X Q b' s> to o p IJ i'O Ồ N LtJ o \o vO Ọ o o 00 00 Vo !° ỈÕ Ồ (O U\ o ƠN VO GO ■o 'O o to to o N z CTQ o 00 o 4^ o 00 U\ TU 5P p> -1 ~' X v0 00 10 Ỉ—> H Q D n (IQ T3 Lr prr Uủ < —> Ì3 ƠQ ~ơ Cl —> * 1— OQ D i•U3< 1^- — > ƠQ & < rCl h —> —> (iQ ƠQ ? r !Z5 ƯQ r -f z ^r* n>> o r-\ ti' ^ S' * £>O' " ^ o “ p ' O- i-L CL g.r,Q •Q Hj r;c o> ĨỊ7* Q*i s ; Ơ0 o £> a o* *T3 r_ -t —1 ƠQ f(3 e- ro I Èr c -J •P ( o “a , ƠQ CO ^ 0' 9*- rr: r~ fro Q' o _D c rfq ỢQ p> fbi rjQ sQ i_££ ' (/> ^ CL ^T rj" r“ Q t-> ƠQ S’ XI Q c/5 H Q c/3 Q 00 H c/5 (/) c/5 2^ DC z H 00 00 Q c r o> H c/5 c po r;c < c> o? o ƠQ c O'* rD>i rt»c n»> a c/} "P Q 00 •p !-j CO c £> c Q H o>> s Ợ* QTQ Cft5)i •P3> i z g < z C3 H J- c S' o | Í •“ 2' OfJQ X z z < -rS>" Ộ _ o o a: z z H ^ < I -rS> C/Q h z 3a= -t 5“ 9' z K S' I z n Ị o ffi >«c 2: < •p r w> z •o fi9 G ọ ặ •>' z o L*J z p o> o _4_ < ƠQ ^ r ƠQ ar z ffi 00 H * H to n o c/5 *c >> H >< X ffi * v •o> a ƠQ 1—] c Q0 ƠQ : CD X n iz: S o5 n i- T O •n» 50; CD- a c ưi *H 'õ — ƠQ < •aj> •Q > H c/5 = q QJ p*r ■J o>i 1-t■Q •[»> " S' r-/ TJ •a>> r r c ƠQ X 05» o p> >< jv,> - n> •o > n>>- 0> > o ow ' co> ^ a 'V p c *-r-i ^ O h ạ> Cd n •— ^ >J-< ■o z 2’ ƠQ Q0 ƠQ § G3 z o c> □ cro s; o o2 z Ui w- HOA XÃ HỘI CHU NGHĨA VIẸT NAM Độc lập - Tụ - Hạnh phúc 0)>, J— o> “Ỉ y^N n ^ > ' w> z CỘNG > rrq ■—1• —> a> n>> r-à “Í PT—» (p o >1 •5> q n» > z DẠI HỌC QUỒC GIA HA NỘI TRUONG ĐẠI HỌC CÔNG NGHỆ to IJ to o '35 u OX) e CJ "«L> Q CXJ Ọ u (—> G o Q '5 ■3 u jo- c3 -C CQ a , H z c «« ',0c13z , ( u ôu- y * > y! J to io I'O to to IO L/l to Ui U\ 10 f—■> [O 'J KJ KJ 10 f—“> to ' -J i to n — f z ƠQ C U\ z o>? Ỉ3 rjrq z r/Q •Ộ o z ƠQ ƠQ I ffQ "a JuP' rE> c.' c Q' r-f p ƠQ 0»- ; 03 •c X“J5 Ơ Í0> G cx o> Ou n»’ zr rD> a’~~*1 •c n»"a nr “ u n I-O ♦■ C o> •o> lJ ƠQ 0 ƠQ •JX)> «” ^ •ro> c p•ÍD> X Cl)? CL P ' t_ c •o> •Q p< u - T5 a LÍ n DJ GO t—3 GO •q ẵT1 0iAJ'T3 P' I Ơ Q 4* ci Q CD> H j ƠQ z n CƠ ƠQ ) X >— Xr— )' JAJ r,0 o u> p 00 H C/Ĩ < —1 • —• c n c? •PC > to r 1o0>CTQ"Tl 0D)>ƠQ*-> o r-O nr- c >? —> rDC >< £)> c/1 •cQ- 2CA , "O Ĩ3 pffQ u- — • )i.( ƠQS’ s •JDV i3> crD/o nj>C P( O n> fu>C r — ►C ' o o> (— +• rti> o n 5* H c/) oo Ồ o oo sỏ 00 \D ; ' 5t ? C l f/Q - g ; f g - _ 9r Ọ n> X.: ~ O ' c ' ƠQ s* CO •i^> f—0 D• r— ^r* ƠQ I_> U5> (;q r- £p ► _» to- o> CO? X —* (fQ i“+• '-V — * rtrrq < ƠQ i ’ a CO o s o ^ ỌC s o 2' O r! X m n z 2! CTQ K) _ c .rn> Q' c f ƠQ c X v: p 3- c3> n t D“ ^r z ƠQ f -• H ƠQ 2> p CL tá ƠQ rS> ữ 3> Í3 ƠQ — > cr p> D o> r~f 0 n ƠQ 00 (—, Qu •p< n ƠQ C/ P o cr o Q' Uủ' rD>X5 P' —> g H-» fTQ n» (IQ t-/ p H «T> rs>' tù ' ƠQ ? •p ĨU- CỊ- z •n» o> *b > (JQ •< o> ƠQ T3 *— > < t-i p> J3- CL E3 •rD> —» ƠQ QJ •n» X f— I3D- z r— ^3> b ~r f— p £»■ t=) T3 i-j n r p UJ> ƠQ o> n»- p p -J r/Q S.Í >< o r-, £ o z ẩ SO 00 c r f—^ rD> r— (IQ 13 (D> D* ZD o> p> D —{ cr tì' fD> r-f Li GO ft)>* 00 so On Ui •p CD>? ÍIỈ , _* oo —• to to r- vO oo sO On z to GO to ƠQ UJ \] 10 10 s> ’3 10 4^ _o Ồ QO ■^o * ỢzQ o OJ Ọ so to £ •?s> TI3 i'O Lti 10 ƠN oo c to to K) o U i r/rQ o to ri)>? H H IO - Q al ổ Ơ ^ E c Q Qffc i fl s n z CD ■Ĩ - 0 i z < p II Ổ 2' (10 a n z z g n z •?D> £ n z < a H Oh d o I' C fp hH o z aỌ ^ d S' O fl !— Li k-4z GJ n z o n o> CTQ ft5> 01 § z E ' ZC u c ^ -5 z p: u a ~ Ọ 9' ^ |- y ff- f;o - • z; o z C0>- cr —< p> C' •o> Q' 7Q &5>i a H o 2' Ổ I' o TO LLc c | —1 i i § n z 9* o>> z z H rD> n c n p o cr ?B>rDu o>- “a •p> »-> ƠQ a» c/3 > X c Ọ~ S ^' ;q z Ơ £»9*o>_ _ g >! c r fro ■T3 CO s.rio 'S , •£>< o p* p o so oo — > p> i-s' < O’ p" T3 V o> o 5’ —> ƠQ ƠQ Ồ •ạ — ) to o Ọ so oo sO X o o K) CO P' Q> D n CƠ d Lh Ồ UJ Ợ c X * l-< ƯQ eo> ƠQ 1—* r Ị a •r— CO-'•"' '5 o ẸT- o C ' o cx z c ặ i = cl c u - -a> o> Ou -n» I Q' ~ b fR o U\ o to Ọ \o z ẻ o z JQ o ụ D£ Ẽ ’5 -T5 "O Q C ỹ J > Q Cũ o CQ o > H < o ò c/j ừì o -cạ p U Qz 11o J (0— -5 HẠ o z D ÍJ p •5 C - r : ‘ C3 H o 'Q r~ D 'I O' Q ề Dí) o s I ặQ co ồtì -D c ‘ C3 H g*o '5 o KS5 n n o n ’ n ‘O Tj ro Ọ C «L) (—■ p (-' )CỢ > ỉ' *■X1 o>? < £>< r~ to o IO to o to Ui to vO G X z z o 00 _l \o sO 00 Ị D a ta p - •> - -a V ie t n a m •f?> '£I-J '< • o P ' - • cro - p jy p> D —1 £ , CO p ƠQ Er ^ p § ” c> c s i ?■ E a 'E c- fJQ •pj p: X) g 03 & -S5 o L*J On Ồ \) o ■ — UJ o o L /1 00 o o VO _ i _ _ t a VO oo to 'O GO Qi T5 o Ễ: 3- tạ3 f—f _ Sc >t - D t >—■ J '-c *&3> ri>> r“f 4^ z i = ; a» ƠQ ■ •a» o a o+ o r, p - ^ X o £> C- *-< ’ I > ~ -c o o !_) f—+■ H o a>> o X q- •&3> o ^ c d r •p p Ob T 0>‘ : X c> o> ^ (5 go> ffQ -ỹ Q 00 — 00 z H c/5 ? ffQ o>? z crq < rD>i PD( Hj ƠQ c;? JD r-f- sa; x: z c c/5 < n» > s fjc 03 p= ' l- J r-^ i 5' ° cH 3( P ƠQ n c ƠQ £ủ' fò> ■p ■o> H r~ n» z rp p S ' CO- o* ^>' o H CL =*■ c- •p> í—K "CJ* Sp H o> o Í3o^ D" r“ Q h3 Cị o>' S 09 _ z IQ _ Vo oo \0 n -g< Si fa, ĨTỈ ^ "D — •P( o S'- ƠQ s: fD>- o> "X) H Q H c/5 t/5 •p I—1 H c/j z o» —> z Cũ< r/Q H •-Í d Q' H ƠQ ƠQ ■Ộ n £U' to —* 1—* w & fĩ)>í a" rS>* c/5 ffQ • ỉ Ỉ3 < ^ S ' fJQ ?"• o», ^ C/D CO ^3- T S3 < ’ < -o> o c/5 o r- f/Q ZJ* o -p < p< C3 *—f- cr n)> “2 - 2^ -o ƠQ ' 00 rD>? q - ffQ ƠQ » CL •c o ƠQ a Ọ* po< Õ o X B ^ T3 X r i' ^_> o so o OJ , * 'VO 00 "5 tT ƠQ CO' -rG - -• —' -a» r> c>— Q_ H p o GO VO - ẽ2 o •p s 00 rio S / O < c ổ > ^ "O too>"Zj -5 ^T- Q cn (— — UJ \) C O 4^ Ồ v ~ H c/i Q c/3 I I1 f i &D» LTt p> Ọ Cf c *2 — ƠQ H c/5 c ^ CO c -a» £)>; Cl £> õ to o ' ’“ ■ n» o p p c ƠQ B' &>' b ~ a a»: ^ '2 IO IO 'J l to H H o> OJ a to o o »—3 o oo ^ ưi 00 •cp> or— o> (O ’ “S B> ỹ r^ ‘ 2- o » ■B3 I I o> c Q- ■?B> 3 Ọ §’■ Ki ■o>“ • o>-Ọ _ Q- o i z0 Ơ f—D >? > cQa < z to o *—■* Ọ I I—f to o to L*-> Õ CJQ oo -J o \ - U\ •Ọ o o ‘— * _ < 'J\ p D _ OJ OJ L /i u~\ to p>> o 00 o 00 —* VO 00 to to o to to t'O o pp _ IO o to to ƠQ I— o* Q>? p ' _ * IO U\ oc o LỊUI.I Ị ol t J IO ƠQ r— n O' •o> o z P3 ^T* CTQ Q ƠQ H ƠQ —5 •> r~ c - z •o > I o c — n H _i c S' (5 r;c X CO' Ơ h7H >-U Ọ > ƠC ặ o z a- o H c 2' O rẩ I-U c Ọ Dh M-t ễ g 9' O rg i s Q> !->1 ' — c Q2 ƠC Ơ X o o z z ' rfo ị o Ơ < E -0 > o n ặ Ọ 9' O rg Sc z : z H n z ể sc D < z < ^ rS> I— pp nn o z _ < z < -n> X g' — o c o> z o u z u X Q z ỒỊ) u s