• 沒有找到結果。

以加權隨機子空間法為基礎之多重分類器系統

N/A
N/A
Info
下載
Protected

Academic year: 2021

Share "以加權隨機子空間法為基礎之多重分類器系統"

Copied!
63
0
0

加載中.... (立即查看全文)

立即下載 ( 63 頁 )

全文

(1)ಃ΋ക ᆣፕ ಃ΋࿯ ࣴ‫୏ز‬ᐒ ଯᆢࡋၗ਑ӧ౜ჴғࢲύቶ‫ݱ‬Ӧ೏٬ҔǴ໺಍‫ޑ‬ϩᜪ‫מ‬ѯύ೿ࢂӃଷ೛‫ڀ‬Ԗ ‫୼ى‬ၗ਑ёа٬ҔٰࡌᄬϩᜪᏔǴฅԶଯᆢࡋၗ਑‫܌‬ሡ‫૽ޑ‬ግၗ਑ঁኧК΋૓ϩ ᜪၗ਑ाٰ‫ޑ‬ӭр೚ӭǴӢԜத཮р౜ၗ਑ঁኧό‫ޑى‬௃‫ݩ‬ǶќѦᗨฅၗ਑‫ૈ܌‬ ග‫ૻޑٮ‬৲ຫӭёа‫׳‬ᔅշၗ਑‫ޑ‬ϩᜪǴՠҗ‫ܭ‬ᆢࡋόᘐቚуǴԶࡌᄬϩᜪᏔ‫܌‬ ሡၗ਑ໆ࣬ჹλ‫ޑ‬௃‫׎‬Ǵ‫ࢂܭ‬྽ᆢࡋၨଯਔǴ߾ϩᜪᏔୖኧ՗ी‫ޑ‬ᆒዴࡋ཮Πफ़ Ъόճ‫ܭ‬ϩᜪǴԜ΋‫ٿ‬ᜤ௃‫׎‬೷ԋ‫܌‬ᒏ‫ ޑ‬Hughes phenomenon (Hughes, 1968)(კ 1-1)౜ຝǶ. MEAN RECOGNITION ACCURACY. 0.75. m=. 0.70. 1000. 0.65 500. 0.60 200 50. 0.55. 100. 20 10 5 m=2. 0.50 1. 2. 5 10 20 200 50 100 MEASUREMENT COMPLEXITY n (Total Discrete Values). 500. 1000. კ 1-1Hughes phenomenon(Hughes, 1968) ! ໺಍ϩᜪБ‫ݤ‬ύჹ‫ܭ‬೭௃‫ݩ‬ೀ౛‫ޑ‬БԄࢂᙖҗᕭλᆢࡋኧٰ෧ϿϩᜪᏔ೛ ीਔ‫ޑ‬ፄᚇࡋǴ΋૓த٬Ҕ‫ٿޑ‬ᅿБ‫ݤ‬Ǵ੝ቻᒧ‫(ڗ‬feature selection)‫܈‬੝ቻ๧‫ڗ‬ (feature extraction) (Fukunaga, 1990)फ़եၗ਑ᆢࡋϐࡕǴӆаൂ΋ϩᜪᏔٰࡌᄬϩ. 1.

(2) ᜪ،฼‫ڄ‬ኧ(decision function)Ƕკ 2-2 ࣁଞჹଯᆢࡋၗ਑‫ޑ‬ϩᜪᏔ೛ीٰࡌᄬϩᜪ Ꮤ‫ࢬޑ‬ำკǶ ! ଯᆢࡋ૽ግၗ਑! ੝ቻᒧ‫!ڗ‬. ଯᆢࡋෳ၂ၗ਑!. ੝ቻ๧‫!ڗ‬. ϩᜪᏔ! ϩᜪਏૈຑ՗! ! კ 1-2 ଯᆢࡋၗ਑‫ޑ‬ϩᜪᏔ೛ीࢬำ. வDietterich(2000)ࣴ‫ز‬ว౜Ǵᗨฅൂ΋ϩᜪᏔ཮ౢғόᒱ‫ޑ‬ϩᜪ҅ዴ౗Ǵՠ όӕ‫ޑ‬ϩᜪᏔϩᜪࡕǴϩᜪᒱᇤ‫ޑ‬ኬҁࠅόᅰ࣬ӕǴ೭ཀ‫ښ‬๱όӕϩᜪᏔ۶Ԝϐ ໔ёૈወᙒ๱ϕံ‫ޑ‬ၗૻǴ‫ߡࢂܭ‬ёаճҔ೭٤ၗૻٰуа‫ׯ‬຾ϩᜪᏔ‫ޑ‬ϩᜪ҅ ዴ౗Ƕ೭ਔջё٬Ҕӭख़ϩᜪᏔ‫ׯٰೌמޑ‬๓ൂ΋ϩᜪᏔό‫ޑى‬ӦБ ǶԶ bagging(Breiman, 1996)ջࢂӭख़ϩᜪᏔ‫מ‬ѯϐ΃ǴЬाࢂ࿶җჹচ‫૽ޑۈ‬ግኬҁ ໣ᒿᐒख़ፄ‫ڗܜ‬ԋኧঁ૽ግኬҁ໣ǴฅࡕճҔ೭٤૽ግኬҁ໣ϩձ૽ግϩᜪᏔǴ നࡕӆ่ӝ೭٤ϩᜪᏔ‫ޑ‬ϩᜪ฼ౣ଺рനಖ،฼Ƕ྽ኬҁኧᆶᆢࡋኧ࣬฻ਔӧጕ ‫୔܄‬ձϩ‫(݋‬linear discriminant analysis, LDA)‫ޑ‬ϩᜪᏔԖܴᡉ‫ޑ‬ਏૈ‫ׯ‬๓(Fisher, 1936)ǶќѦҗHo(1998)‫܌‬ගр‫ޑ‬ᒿᐒη‫ޜ‬໔‫(ݤ‬random subspace space, RSM)Ψ᛾ ܴрǴӧၗ਑р౜λኬҁଯᆢࡋ௃‫ݩ‬Ǵჹᆢࡋ‫ڗܜ‬٠ࡌᄬηᆢࡋኬҁ໣ख़ཥ૽ግ ӭঁϩᜪᏔӆ଺่ӝૈှ،ኬҁኧό‫ޑى‬௃‫ݩ‬Ƕᒿࡕ‫׳‬җ(Kuo, Hsieh, Liu, & Chao, 2005)‫܌‬ගрǴу៾ᒿᐒη‫ޜ‬໔‫(ݤ‬weighted random subspace methods, WRSM)Ǵа RSMࣁ୷ᘵǴଷ೛η‫ޜ‬໔ᆢࡋϐᒧ‫ࢌܭ୷ࢂڗ‬ϩଛǴЪӧᒧ‫ڗ‬ᆢࡋӕਔаਡѳྖ. 2.

(3) ϯ‫(ݤ‬kernel smoothing)ٰ՗ी٠‫׳‬ཥ‫ځ‬ϩѲ‫ڄ‬ኧǴനࡕӆ่ӝ‫؂‬ԛᒧ‫ڗ‬ᆢࡋϐη ᆢࡋኬҁ໣‫૽܌‬ግрٰ‫ޑ‬ϩᜪᏔǴӵԜ‫ڗܜ‬ᆢࡋϐБ‫ૈݤ‬ှ،RSMѝ‫ۓڰ‬ჹᆢࡋ ‫ڗܜ‬rঁ‫ޑ‬લᗺǶҁࣴ‫ز‬යఈૈᙖҗ่ӝ೭٤‫מ‬ѯှ،ൂ΋ϩᜪᏔϷλኬҁ‫ޑ‬ୢ ᚒǴ‫ࢂܭ‬ගрbaggingᆶWRSMϐ่ӝ٠ࡌᄬќ΋ӭख़ϩᜪᏔǴٰೀ౛Ҟ߻ӧଯᆢ ࡋၗ਑ϩᜪ྽ύ‫܌‬य़ᖏ‫ޑ‬ୢᚒǶ! !. ಃΒ࿯ ࣴ‫ز‬Ҟ‫ޑ‬ ҁࣴ‫ز‬ஒ‫ ؼׯ‬bagging ᆶ WRSMǴଞჹ٬ҔচБ‫ޑݤ‬લᗺϷࡑှ،‫ޑ‬ୢᚒǴ ගрԜཥ‫ޑ‬ӭख़ϩᜪᏔ‫س‬಍ǴԶჹ‫ޑزࣴܭ‬ЬाҞ‫ޑ‬ӵΠ‫܌‬ҢǺ 1.. ගр bagging ᆶ WRSM ่ӝϐཥӭख़ϩᜪᏔǴࣴ‫ز‬ᆶᙑБ‫ ݤ‬bagging ᆶ WRSM ਏૈϐৡ౦Ƕ. 2.. ૸ፕόӕୖኧ(ӵϩᜪᏔঁኧ)ჹϩᜪਏૈ‫ޑ‬ቹៜᆶᡂϯǶ. ! ҁࣴ‫ز‬ύ‫܌‬௦Ҕ‫ޑ‬ଯӀ᛼ᇿෳቹႽаϷ௲‫ػ‬ෳᡍၗ਑ճҔଯථϩᜪᏔ (quadratic Gaussian density classifier, qdc)ǵk ന߈ᎃϩᜪᏔ(k nearest neighbors classifier, knnc)‫ک‬ЍኖӛໆᐒϩᜪᏔ(support vector machine classifier, svc)Οᅿϩ ᜪᏔǴ่ӝ bagging ᆶ WRSM ᔈҔӧӭख़ϩᜪᏔ΢ٰᡍ᛾ҁࣴ‫܌ز‬ගБ‫ޑݤ‬ёҔ ‫܄‬Ǵനࡕ่݀ᡉҢǴҁࣴ‫ز‬ϐ‫܌‬ගБ‫ݤ‬ёаග‫ٮ‬΋ঁၨ໺಍ӭख़ϩᜪᏔӧࢌ٤ၗ ਑ಔӝΠԖ‫׳‬٫‫ޑ‬ϩᜪ҅ዴ౗Ƕ! !. ಃΟ࿯!಄ဦᇥܴ ҁ࿯ஒϟಏҁࣴ‫܌ز‬٬Ҕ‫ډ‬಄ဦǴ٠଺ᙁൂᇥܴǶ. 3.

(4) ૽ግኬҁ໣: D {( xi , yi ) | 1 d i d N }, xi  X  S , yi {1,..., L}  C , i 1,..., N x :ൂ΋ኬҁᗺ. iǺ૽ግኬҁᗺ઩Їࡰ኱ y :ኬҁᗺ x ჹᔈ‫ޑ‬ᜪձ. jǺᜪձ‫ޑ‬઩Їࡰ኱ S  ƒ p :ኬҁ‫ޑ‬চ‫ޜۈ‬໔. p : ኬҁᆢࡋኧ X. ( x1 , x2 ,...x N ) :җ N ঁኬҁᗺಔԋ‫ޑ‬ၗ਑໣. NǺ૽ግኬҁᗺኧ CǺϩᜪ،฼‫ޑ‬ှ໣ӝ LǺᜪձኧ hǺϩᜪᏔ rǺ‫ۓڰ‬η‫ޜ‬໔ᆢࡋኧǴΨ൩ࢂ੝ቻኧ F  ƒ' : ੝ቻ‫ޜ‬໔. A : ᙯඤંତ Pj : ᜪձ j ‫ޑ‬Ӄᡍᐒ౗. BǺϩᜪᏔঁኧ bǺϩᜪᏔঁኧ‫ޑ‬઩Їࡰ኱ h final : S o C നಖ،฼ǴջϩᜪᏔ. U : ϟ‫ ܭ‬1 ‫ ډ‬p ໔ϐᚆණ֡΋ϩѲ(discrete uniform distribution) R : η‫ޜ‬໔ᆢࡋख़ाำࡋϩѲ(importance distribution of subspace dimensionality) WǺ੝ቻख़ाำࡋϩѲ(importance distribution of features) B0 Ǻ߃‫ۈ‬໘ࢤҔ‫ܭ‬՗ी R0 ‫ޑ‬ϩᜪᏔঁኧ. 4.

(5) R0 ǺR ϐ߃‫ۈ‬ϩѲ. KǺਡ‫ڄ‬ኧ(kernel function) !. 5.

(6) ಃΒക Ў᝘௖૸ ӧҁക྽ύஒჹࡌᄬԜཥ‫ޑ‬ӭख़ϩᜪᏔ‫س‬಍բ΋٤Ў᝘‫ޑ‬௖૸Ƕ. ಃ΋࿯!੝ቻᒧ‫!ڗ‬ ੝ቻᒧ‫ޑڗ‬Ҟ኱ᙁൂٰᇥ൩ࢂ‫ޔ‬ௗவኬҁচԖ‫ޑ‬੝ቻύᒧрჹϩᜪਏ݀ૈ Ԗշ੻‫ޑ‬೽ϩ੝ቻǴ٬‫ځ‬ϩᜪ҅ዴ౗ૈ୼ၲ‫ډ‬നଯॶǶ೭٤᠘ձૈΚၨӳ‫ޑ‬੝ ቻǴόՠૈ୼ᙁϯϩᜪᏔ‫ޑ‬ၮᆉǴΨёаᕕှԜϩᜪୢᚒ‫ޑ‬Ӣ݀ᜢ߯Ƕ೭္ჹΟ ᅿ୷ҁ੝ቻᒧ‫ڗ‬Б‫ ݤ‬ForwardǵBackwardǵIndividual ຾Չᙁൂ‫ޑ‬௖૸ǶForward ࢂ ΋ᅿӛ߻೴‫؁‬ཛྷ઩‫ޑ‬ᄽᆉ‫ݤ‬Ǵ२Ӄ‫ঁ؂ע‬ᆢࡋঁձ‫ޑ‬ϩᜪ҅ዴ౗ीᆉрٰǴௗΠ ٰ‫ע‬നӳ‫ޑ‬ᆢࡋᒧ‫ڗ‬рٰǴௗΠٰ‫ע‬ഭΠ‫ޑ‬ᆢࡋ೿ᆶ೏ᒧ‫ڗ‬р‫ޑ‬ᆢࡋ଺ಔӝǴी ᆉঁձ‫ޑ‬ϩᜪ҅ዴ౗ǴӵԜϸᙟ‫ޑ‬ीᆉᆶᒧ‫ډޔڗ‬ന٫‫ޑ‬ಔӝр౜ࣁ ЗǶ Backward ࢂ΋ᅿӛࡕ‫ޑ‬೴‫؁‬ཛྷ൨‫ޑ‬ᄽᆉ‫ݤ‬ǴӃ‫܌ע‬Ԗ‫ޑ‬ᆢࡋ࣮ԋࢂ΋ᅿಔӝǴ ௗΠٰ‫؂‬ԛ೿‫ע‬΋ঁᆢࡋ‫ڗ‬рǴीᆉഭΠಔӝ‫ޑ‬ϩᜪ҅ዴ౗Ǵᒧ‫ڗ‬ഭΠಔӝ҅ዴ ౗നଯ‫ޑ‬٠ख़ፄϐ߻‫؁ޑ‬ᡯǴϸᙟ୺Չ‫ډפډޔ‬ϩᜪ҅ዴ౗നଯ‫ޑ‬ಔ ӝǶ Individual ࢂ‫܌ע‬Ԗ‫ޑ‬ᆢࡋ೿࣮ԋࢂᐱҥ‫ޑ‬ǴӃीᆉрঁձ‫ޑ‬ϩᜪ҅ዴ౗Ǵӆ٩ Ᏽϩᜪ҅ዴ౗‫ޑ‬ଯեբ௨‫ׇ‬Ǵᒧ‫ڗ‬ᆢࡋਔ൩җϩᜪ҅ዴ౗ၨଯ‫ޑ‬ᆢࡋ‫ޔ‬ௗӃՉᒧ ‫ڗ‬ǴόႽ‫ځ‬Ѭᄽᆉ‫ݤ‬ԖԵቾಔӝ‫ޑ‬ϩᜪ҅ዴ౗Ƕ. ಃΒ࿯!੝ቻ๧‫!ڗ‬ ੝ቻ๧‫ڗ‬٣ჴ΢ࢂ੝ቻᒧ‫ۯޑڗ‬՜ǴനЬा‫ޑ‬Ҟ‫׆ࢂޑ‬ఈஒၗ਑ဂૈ୼җ‫܌‬ ӧ‫ޑ‬ၨଯᆢࡋ‫ޜ‬໔ύ‫׫‬ቹ‫ډ‬ၨեᆢࡋ‫ޜޑ‬໔ύǴ‫ځ‬ᆶ੝ቻᒧ‫ڗ‬നε‫ޑ‬όӕӧ‫܌ܭ‬ ࡷᒧр‫׫ޑ‬ቹ‫ޜ‬໔୷ۭёаҗচҁ‫ޑ‬੝ቻ୷ۭ࿶ၸጕ‫ߚࢂޣ܈܄‬ጕ‫ޑ܄‬ၮᆉٰ. 6.

(7) ಔӝԶԋǶଷ೛ஒ‫؂‬΋฽ၗ਑‫ޑ‬Ӛঁ੝ቻຎࣁၗ਑‫܌‬ӧ‫ޑ‬০኱Ǵٗሶ൩ёаஒӄ ೽‫ޑ‬ၗ਑ຎࣁ΋ဂϩѲӧଯᆢࡋ‫ޜ‬໔ύ‫ޑ‬ၗ਑ᗺǴԶ‫؂‬฽ၗ਑‫ޑ‬੝ቻኧҞ൩ёа ຎࣁ၀฽ၗ਑‫ޑ‬ᆢࡋǶҁࣴ‫ࡌ܌ز‬ᄬ੝ቻख़ाำࡋϩѲ‫ڄ‬ኧǴࢂҗϩѲ‫ڄ‬ኧύᒧ ‫ڗ‬੝ቻǴ‫ځ‬ύϐ΃ࢂ٬Ҕጕ‫୔܄‬ձϩ‫݋‬ᜪձϩᚆໆ(class separability)ٰբࣁ੝ቻ ‫ॶ៾ޑ‬ǴԶӧ຾Չ୔ձϩ‫݋‬੝ቻ๧‫(ڗ‬discriminate analysis feature extraction, DAFE) ਔǴᜪಔϩᚆໆ‫(߾ྗޑ‬criterion)ࢂҗಔϣϩණંତ(within-class scatter matrix)ǵಔ ໔ϩණંତ(between-class scatter matrix)Ϸషӝϩණંତ(mixture scatter matrix)ٰ ‫׎‬ԋǶ‫ځ‬ύǴಔϣϩණંତᆶಔ໔ϩණંତϩձ‫ۓ‬ကӵΠȐFukunaga, 1990ȑǺ ಔϣϩණંତǺ S wDA. L. L. ¦ Pj 6 j. ¦ P (X  M j. j 1. j. )( X  M j )T. (2-1). j 1. ΢Ԅ‫ ޑ‬L ߄ҢᜪಔᕴኧǴ Pj Ϸ M j ϩձࣁӃᡍᐒ౗(prior probability)ϷӚᜪಔ ‫ޑ‬ѳ֡ኧӛໆ(mean vector of the class j ) ಔ໔ϩණંତǺ S bDA. L. ¦ P (M j. j.  M 0 )( M j  M 0 )T. (2-2). j 1. ‫ځ‬ύ M 0 ж߄షӝϩଛ‫ޑ‬යఈӛໆǴԶЪϦԄ‫ۓ‬ကӵΠǺ L. M0. ¦P M j. (2-3). j. j 1. চ‫ޜۈ‬໔ X ࿶җ΋ঁᙯඤંତ A ࡕǴ٬ள X ёаᙯඤ‫ډ‬ཥ‫ޑ‬੝ቻ‫ޜ‬໔ Y Ǵ٬ ளԖ Y. AT X ‫ޑ‬ᜢ߯Ǵ߾җԜё‫ޕ‬Ǻ S wY. AT S wX A. Ъ. S bY. AT S bX A. ᜪಔϩᚆໆ‫ۓߡ߾ྗޑ‬ကԋǺಔ໔ϩᚆໆཇεǴಔϣϩᚆໆཇλ߾ཇճ‫ܭ‬ϩ ᜪǴӢԜǴᜪಔϩᚆໆ‫߾ྗޑ‬ϦԄϷന٫ϯှߡ‫ۓ‬ကӵΠǺ J DAFE ( p). 1 trace( S wY S bY ). 7. (2-4).

(8) ࿶җന٫ϯၸำǴவ J DAFE ύᒧ‫ ߻ڗ‬p ঁၨε‫ޑ‬੝ቻॶ(eigenvalues)‫܌‬ჹᔈ‫ޑ‬ ੝ቻӛໆ(eigenvectors)ٰ‫׎‬ԋᙯඤંତ ApDAFE Ǵҗ΢ॊᜪಔϩᚆໆ‫߾ྗޑ‬ёளന٫ ‫ޑ‬੝ቻ(optimal features)ǶԶӧҁࣴ‫ز‬྽ύаൂ΋੝ቻ‫ޑ‬ጕ‫୔܄‬ձϩ‫݋‬ᜪಔϩᚆໆ ٰբࣁ੝ቻᒧ‫ޑڗ‬చҹǶ !. ಃΟ࿯ ϩᜪᏔ ҁ࿯ϟಏҁࣴ‫܌ز‬٬ҔϐΟᅿϩᜪᏔǴ၁ಒӵΠ‫܌‬Ң:. ൘ǵʳଯථϩᜪᏔ ଷ೛Ԗ y1 , y 2 ‫ঁٿ‬ᜪձǴЪࣣࣁଯථϩѲ(Gaussian distribution)ǴϩѲϦԄ߄ ҢӵΠǺ 1. N X (M j , 6 j ). n 2. (2S ) | 6 j |. 1 2. 1 exp( ( X  M j ) T 6 j1 ( X  M j )), 2. j. 1,2. (2-5). ‫ځ‬ύ X ࢂᢀჸӛໆǴ M j ߄ಃ j ঁᜪձ‫ޑ‬ѳ֡ኧǴ 6 j ߄ಃ j ঁᜪձ‫ޑ‬Ӆᡂኧ ંତǶଷ೛‫ঁٿ‬ᜪձ‫ޑ‬Ӄᡍᐒ౗(prior probability)ࣁ P1 ǵ P2 Ǵచҹᐒ౗ஏࡋ‫ڄ‬ኧ p1 ( X ). p ( j1 | X ) ǵ p 2 ( X ). p ( j 2 | X ) Ǵ h( X ) ߄Ң‫ن‬М୔ձ‫ڄ‬ኧȐBayes discriminant. functionȑϦԄ߄ҢӵΠǺ h( X ).  ln p1 ( X )  ln p 2 ( X ). (2-6). ྽ h( X ) ε‫ ܭ‬0Ǵ߾ X ղ‫ۓ‬ឦ‫ ܭ‬j 2 Ǵց߾ղ‫ۓ‬ឦ‫ ܭ‬j1 Ƕ Ӣࣁ j1 , j 2 ࣣࣁதᄊϩѲǴ‫܌‬а h( X ).  ln p1 ( X )  ln p 2 ( X ). 1 1 1 |6 | ( X  M 1 ) T 611 ( X  M 1 )  ( X  M 2 ) T 6 21 ( X  M 2 )  ln 1 2 2 2 62. (2-7). ёаว౜྽ 61 z 6 2 ਔǴ h( X ) ࣁΒԛԄ(quadratic form)Ǵ߾ԜϩᜪᏔΞᆀࣁ. 8.

(9) quadratic discriminant classifier (qdc)Ƕ. ມǵʳk ന߈ᎃϩᜪᏔ ଷ೛ϩѲ‫ޑ‬ᐒ౗ஏࡋ‫ڄ‬ኧ‫ޑ‬՗ीԄࣁ Pˆ ( x y j ). k 1 N j v( x). (2-8). ‫ځ‬ύ k ߄Ңനௗ߈ x ‫߈ޑ‬ᎃঁኧǴ v (x) ߄ҢᡏᑈȐvolumeȑǴ N j ࣁᜪձ j ‫ঁޑ‬ ኧǶа‫ٿ‬ᜪձ‫ٯޑ‬ηٰᇥǴଷ೛୔ձ‫ڄ‬ኧ h( X ) ࣁ  ln pˆ 1 ( X )  ln pˆ 2 ( X ). h( X ). k1  1 N v (X )  ln 1 1 k2  1 N2v2 ( X ). n ln.  ln. d2 (X. ( 2) k 2 NN. ,X). d1 ( X. (1) k1 NN. ,X). ( k1  1) N 2 v 2 ( X ) ( k 2  1) N 1 v1 ( X ). 1/ 2.  ln. (k1  1) N 2 | 6 2 | (k 2  1) N 1 | 61 |1 / 2. y1 ! . t. (2-9). y2. ‫ځ‬ύ v j. n 2. S n / 2 * 1 (  1) | 6 j |1 / 2 d nj , d 2j (Y , X ) (Y  X ) T 6 j 1 (Y  X ). ୖǵʳЍኖӛໆᐒϩᜪᏔ ЍኖӛໆᐒϩᜪᏔ(support vector machine classifier, svc)ࢂ΋ᅿᐒᏔᏢಞ‫ݤ‬Ҕ ٰೀ౛ёϩᚆ‫ޑ‬ၗ਑ǴѬ‫ޑ‬ϩᜪ‫מ‬ѯࢂ߈ԃٰ‫ݙڙ‬Ҟ‫زࣴޑ‬ЬᚒǶ౛ፕࢂ୷‫่ܭ‬ ᄬ॥ᓀനλϯ(structural risk minimization)‫ۺཷޑ‬Ƕӧ೚ӭᔈҔύǴЍኖӛໆᐒ౛ ፕК໺಍Ꮲಞᐒ‫ڋ‬Ԗ‫׳‬ଯ‫ޑ‬ਏૈ߄౜ǴЪӧှ،ϩᜪୢᚒ΢ρ࿶ࢂமԶԖΚ‫ޑ‬π ‫ڀ‬ϐ΋ǶЍኖӛໆᐒ౛ፕஒᒡΕၗ਑ࢀ৔Կଯᆢࡋ੝ቻ‫ޜ‬໔Ъ൨‫פ‬ёϩᚆ 2 ঁᜪ ձ‫ޜޑ‬໔ύǴ‫ڀ‬Ԗനεᜐࣚ(margin)‫ޑ‬ёϩᚆຬѳय़(hyperplane)Ƕനεϯᜐࣚࢂ. 9.

(10) ΒԛೕჄ(quadratic programming)ୢᚒǶૈ࿶җ Lagrangian multipliers ᙯᡂԋჹଽ ਱Ԅ‫ޑ‬ୢᚒٰှ،Ƕsvc ൨‫פ‬ന٫ѳय़ࢂճҔ੝ቻ‫ޜ‬໔ύ‫ڄ‬ኧ‫ޑ‬ᗺᑈᆀࣁ kernelǶ ന٫ѳय़ࢂҗϿኧ‫ޑ‬ᒡΕᗺಔӝԶԋǴᆀࣁЍኖӛໆǴ‫ځ‬চ‫ۓۈ‬ကӵΠǺ N 1 min wT ˜ w  O ¦ [ i 2 i 1. subject. to. y i ( wT ˜ I ( xi )  b ) t 1  [ i ,. [ i t 0.. (2-10) i 1,..., N. ӧԜbࢂୃᇤǶӧԜ૽ግኬҁ xi ࿶җ‫ڄ‬ኧ I ࢀ৔Կ΋ঁၨଯᆢࡋ‫ޜ‬໔‫ޜޑ‬໔ǶԜ‫ޜ‬ ໔ёനεϯᜐࣚǶ w ࢂёϩᚆຬѳय़ӛໆǴ [ i ࢂϩᜪᒱᇤ‫ޑ‬৒೚ໆǴ O ࢂж߄ᒱ ᇤ‫ޑ‬ፓ࿯ୖኧЪࢂதኧǴԜୖኧࢂsvc୤΋ёаፓ᏾‫ޑ‬ᡂ୏Ԝୖኧૈѳᑽᜐࣚελ ϷϢ೚ϩᜪᒱᇤໆǶ‫פ‬൨ന٫ຬѳय़ǴёஒୢᚒаLagrangianٰှ،Ъஒচୢᚒ ᙯԋჹଽ‫׎‬Ԅ! ! N. max W (D ). ¦D i  i 1. 1 N N ¦¦D iD j yi y jI ( xi )I ( x j ) 2i1 j1. ! !. N. subject. to. ¦yD i. i. !. !. (2-11). 0 d D i d O , i 1,..., N. 0. i 1. ӧԜ D (D1 ,...,D N ), ࢂߚॄLagrange!multipliers‫ޑ‬ӛໆǶKuhn-Tucker‫ۓ‬౛ӧЍኖӛ ໆϩᜪᏔ‫ޑ‬౛ፕύ‫ת‬ᄽख़ा‫فޑ‬ՅǴਥᏵԜ౛ፕाှ،(2-11)‫ޑ‬ୢᚒǴ߾ D ‫ޑ‬ୢᚒ ሡᅈ‫!ى‬ D ( yi ( w ˜ I ( xi )  T )  1  [ i ) 0, (O  D i )[ i. 0,. i 1,..., N !. i 1,..., N !!. !. !. !. !. (2-12)!. !. !. !. (2-13)!. வ೭٤฻Ԅё‫ޕ‬ѝԖӧ(2-12)ߚ႟ॶ‫ ޑ‬D i ࢂᅈ‫ى‬ज़‫ڋ‬Ԅ y i ( w ˜ I ( xi )  T ) t 1  [ i Ъ҅ ॄ࣬฻Ƕၗ਑ᗺჹᔈ D i ! 0 ೏ᆀࣁЍኖӛໆǴёࢂЍኖӛໆӧόёϩᚆ‫ޑ‬௃‫׎‬ਔǴ Ԗ‫ٿ‬ᅿόӕ‫ޑ‬ᜪࠠǶӧ 0  D i  O ჹᔈ‫ޑ‬Ѝኖӛໆᅈ‫ى‬฻Ԅ y i ( w ˜ I ( xi )  T ) 1 Ъ. 10.

(11) [i. 0 Ǵӧ D i. O ‫ޑ‬௃‫׎‬Π߾ [ i όࢂ‫ޑޜ‬Ъჹᔈ‫ޑ‬Ѝኖӛໆόᅈ‫ى‬ज़‫ڋ‬Ԅ. y i ( w ˜ I ( xi )  T ) t 1  [ i ǴԜᜪ‫ޑ‬ЍኖӛໆࣁᒱᇤǶ྽ၗ਑ᗺ xi ჹᔈ D i. 0 ߄Ңϩᜪ. ҅ዴЪૈܴዴ‫ޑ‬ӧ،฼ᜐࣚϩ႖Ǵࡌᄬന٫ϯຬѳय़ w ˜ I ( xi )  T ࢂа! N. w˜. ¦D y I ( x ) ! ! i. i. i. !. !. !. !. !. (2-14)!. i 1. Ъપໆ T ࢂҗ(2-12)‫ޑ‬Kuhn-Tuckerచҹٰ،‫ۓ‬Ƕӧനॶ٫،‫ࡕۓ‬Ǵ‫ځ‬،฼‫ڄ‬ኧ൩ࢂ! N. f ( x) sign ( w ˜ I ( xi )  T ) sign (¦D i yiI ( xi ) ˜ I ( x)  T ) !. !. !. (2-15). i 1. !. ಃѤ࿯ ӭख़ϩᜪᏔᆶ่ӝ฼ౣ! ϩᜪᏔ‫่ޑ‬ӝനЬाࢂ่ӝӭख़ϩᜪᏔϐঁձ‫ޑ‬،฼ǴԶӵՖ่ӝϩᜪᏔ߾ ࢂቹៜӭख़ϩᜪᏔਏૈ‫ޑ‬Ьा᝼ᚒ(Skurichina & Duin, 2002*ǶԶ٬Ҕӭख़ϩᜪၨ ൂ΋ϩᜪᏔ٫‫ޑ‬চӢǴёаவ಍ी΢Ϸीᆉ΢ٰϩ‫݋‬Ƕ಍ी΢‫ޑ‬চӢࢂӢࣁଷӵ ค‫૽ޑ୼ى‬ግኬҁёа૽ግϩᜪᏔǴჹ‫ܭ‬ϩᜪᏔҁ‫ࡌޑي‬ᄬ཮࣬྽όᛙ‫ۓ‬Ǵ຾Զ ೷ԋϩᜪਏ݀཮ԖୃৡǴ‫ࢂܭ‬ЯಔӝӭಔϩᜪᏔջёှ،ԜୢᚒǶӧीᆉБय़ٰ ᇥǴऩӚϩᜪᏔҗှ‫ޜ‬໔‫ޑ‬ᜐࣚ໒‫ۈ‬ཛྷ൨‫؃‬ှǴ཮Ӣࣁ૽ግਔଆ‫ۈ‬ᗺ‫ޑ‬όӕԶள ‫ډ‬όӕ‫ޑ‬ှǴ‫܌‬а཮தวғค‫ݤ‬ᒧ‫ڗ‬፾྽‫૽ۈ߃ޑ‬ግᗺ‫ࢂ܈‬ό‫ޑؼ‬ཛྷ൨БԄǴന ࡕ཮ഐΕ୔ୱှǶӢԜ୷‫ܭ‬೭٤౛җٰᒧ᏷ࡌᄬӭख़ϩᜪᏔǴԶკ 2-1 ߾ࣁ΋૓ ӭख़ϩᜪᏔϐ೛ीࢬำǶ! !. 11.

(12) ϩᜪᏔ‫ޑ‬ౢғ!. ϩᜪᏔ‫ޑ‬ᒧ᏷! ಔӝϩᜪᏔ‫ޑ‬೛ी! ਏૈ‫ޑ‬՗ी! ! კ 2-1 ӭख़ϩᜪᏔ‫ޑ‬೛ीࢬำ(Bunke & Kandel, 2002)! ! ϩᜪᏔ่ӝёаКൂᐱ٬Ҕ΋ᅿϩᜪᏔ‫ޑ‬௃‫ݩ‬ΠԖ‫׳‬ӳ‫ޑ‬ϩᜪ҅ዴ౗Ǵࠅѝ ሡाӆӭ଺΋ᗺᚐѦ‫ޑ‬ၮᆉǴ೭ࢂӢࣁόӕ‫ޑ‬ϩᜪᏔϐ໔ёૈග‫ٮ‬Αϕံ‫ޑ‬ၗ ૻǴӢԜ่ӝኧঁϩᜪᏔჹ‫଺ܭ‬рനࡕ‫ޑ‬،฼ૈԖ‫܌‬շ੻ǶςԖ೚ӭόӕሦୱ‫ޑ‬ ჴᡍ่݀ᡉҢǴ‫܌‬а่ӝϩᜪᏔࢂ΋ঁԖਏගϲϩᜪ҅ዴ౗‫ޑ‬Б‫ݤ‬ǶҞ߻ӧࣴ‫ز‬ ሦୱ΢Ǵӭख़ϩᜪᏔຫٰຫ‫ډڙ‬ኬԄᒣᇡࣴ‫ޑޣز‬ख़ຎǶӢࣁ྽ൂ΋ϩᜪᏔόᛙ ‫ۓ‬ਔǴӭख़ϩᜪᏔёаК٬Ҕൂ΋ϩᜪᏔ‫܌‬ள‫ޑ‬ਏૈ‫׳‬٫Ƕλኬҁଯᆢࡋ‫ޑ‬௃‫׎‬ ཮೷ԋ‫܌‬٬Ҕ‫ൂޑ‬΋ϩᜪᏔ՗ीୃᇤϷόᛙ‫ۓ‬Ǵаӭख़ϩᜪᏔٰ‫ڗ‬жচԖ‫ൂޑ‬΋ ϩᜪᏔБ‫ݤ‬ёаගଯϩᜪᏔ‫ૈ܄ޑ‬Ϸᛙ‫܄ۓ‬ǶਥᏵӃ߻‫(زࣴޑ‬Roli, 2002)Ǵ่ӝ ϩᜪᏔ‫ޑ‬Б‫ݤ‬ёϩࣁаΠΟᅿǺ!!. 1. ‫ׇ‬ӈԄ่ӝ(serial combination) ኧঁϩᜪᏔа‫ׇ‬ӈ‫ޑ‬БԄӼ௨Ǵಃ΋ঁϩᜪᏔ‫཮่݀ޑ‬ԋࣁಃΒঁϩᜪ Ꮤ‫ޑ‬ᒡΕǶӵ݀߻य़‫ޑ‬ϩᜪᏔၶ‫ډ‬ᜤаϩᜪ‫ޑ‬ኬҁǴ߾཮ஒԜኬҁҬҗࡕय़ ‫ޑ‬ϩᜪᏔаόӕ‫ޑ‬ᆢࡋٰϩᜪǶҗԜё‫ޕ‬ǴϩᜪᏔ‫ޑ‬Ӄࡕ໩‫ׇ‬ஒ཮،‫ۓ‬Ԝ‫س‬ ಍‫ޑ‬ਏૈǶ!. 12.

(13) 2. ϩᜪᏔᑼӝ‫܈‬٠ӈԄ่ӝ(classifier fusion or parallel combination) ‫܌‬Ԗ‫ޑ‬ϩᜪᏔӚձჹኬҁ଺ϩᜪǴϩᜪࡕ‫่݀ޑ‬ӆҬҗ่ӝኳಔ (combining module)ٰ᏾ӝǴ٠଺р΋ठ‫ޑ‬،‫ۓ‬Ƕ᏾ӝ‫ޑ‬БԄΞёϩࣁ majority voteǴ Bayesian(productǵsumǵmaxǵminǵmedian ‫ ک‬mean)…฻Ƕ. 3. ୏ᄊϩᜪᏔᒧ‫(ڗ‬dynamic classifier selection) 㝕ঁϩᜪᏔჹኬҁ‫ޑ‬੝ቻ‫ޜ‬໔(feature space)‫ޑ‬όӕ୔ୱ଺ϩᜪǴ྽΋ঁ ཥ‫ޑ‬ኬҁᒡΕਔǴ‫س‬಍཮ႣෳՖᅿϩᜪᏔ‫่݀ޑ‬ωࢂ҅ዴ‫)ޑ‬Ψ೚ࢂ΋ঁ‫܈‬ ኧঁϩᜪᏔ*Ǵ٠଺рനࡕ‫ޑ‬،฼Ƕ! ! ϩᜪᏔ่ӝ฼ౣ(classifier combination scheme)‫ځ‬Ҟ‫ࢂޑ‬ஒӭಔ‫ޑ‬ϩᜪᏔ຾Չ ่ӝ‫୏ޑ‬բǴԵቾჹၗ਑ᗺ X ଺ϩᜪǴԖ j ঁёૈ‫ޑ‬ᜪձ y j - j 1,..., L ǴԶၗ਑ᗺ X ѝ཮೏ࡰࢴ‫ځډ‬ύ‫ޑ‬΋ঁᜪձǶӅԖ B ঁϩᜪᏔǴ‫ঁ؂‬ϩᜪᏔჹԜၗ਑ᗺ X ೿. Ԗόӕ‫ޑ‬ෳໆӛໆ(measurement vector) xb - b 1,..., B ǶӢԜёаள‫ډ‬ᐒ౗ஏࡋ‫ڄ‬ኧ (probability density function) p( xb | y j ) ᆶ Ӄ ᡍ ᐒ ౗ (a priori probability of occurrence) P ( y j ) ǶӧԜҔ‫ډ‬Ьा‫׫‬౻‫(ݤ‬majority vote)‫่ޑ‬ӝ฼ౣǴ่ӝ฼ౣύࣁ Α҅ዴ‫ޑ‬ճҔ‫܌‬Ԗёள‫ޑ‬ၗૻٰ଺р،฼ǴΨѸሡӕਔԵቾ‫܌‬Ԗ‫ޑ‬ᢀෳॶ٠ीᆉ ‫ࡕځ‬ᡍᐒ౗(a posteriori probability) P ( y j | xb ) ǴճҔࡕᡍᐒ౗ౢғΒ຾Տ‫ޑ‬ᒡр! L. ' jb. ­1 if P( y j|y b ) max P( y j|xb ) j 1 ® ¯0 otherwise. ਥᏵ΢ԄǴଷ೛ࣁ equal priorsǴёаள‫ډ‬Ǻ! ࡰࢴ! X o y j !!. 13.

(14) B. ӵ݀! ¦ ' jb b 1. B. L. max ¦ ' jb j 1. (2-16)!. b 1. ಃϖ࿯ Bagging Bagging ࢂҗ!Breiman ӧ 1996 ԃගрǴ౛ፕࢂჹচҁ‫૽ޑ‬ግၗ਑໣ౢғӭಔ ཥ‫૽ޑ‬ግၗ਑໣ǴճҔ೭٤ၗ਑໣‫ޑ‬ৡ౦‫܄‬ϩձ૽ግӭঁϩᜪᏔǴӆճҔ೭٤ϩ ᜪᏔӕਔჹኬҁ຾ՉϩᜪǴ‫܌‬ள่݀аЬा‫׫‬౻‫ٰݤ‬،‫ۓ‬ǴԜ‫ݤ‬ёගଯόᛙ‫ۓ‬ᄽ ᆉ‫ޑݤ‬ϩᜪ҅ዴ౗ǶਥᏵ(Breiman, 1996)‫ ޑ‬bagging ‫ޑ‬ϐᄽᆉ‫ݤ‬Ǵ‫ۓځ‬ကӵΠǺ! ! [bagging ᄽᆉ‫]ݤ‬ 1. Repeat for b 1,2,..., B ʳ (a)Take a bootstrap replicate X b of the training data set X 2. Combine classifiers C b ( X ), b 1,2,..., B by majority voting (the most often predicted label) to a final decision rule E ( x) arg max ¦ G C y{1,..., L}. b. b. ( x ), y. (2-17). where G i , j is the Kronecker symbol, and y 1,2,..., L is a decision (class label) of the classifier for two-class problem.. !. Bagging ‫ޑ‬ϩᜪਏ݀ࢂຎ૽ግኬҁኧελ‫ک‬ኬҁᆢࡋኧԶ‫ۓ‬Ǵӧ୷ྗϩᜪᏔ‫ޑ‬. λኬҁ‫܄‬፦Ǵ྽ϩᜪᏔࢂߚሀ෧ᏢಞԔጕ٠٬Ҕ LDA ਔǴbagging ‫ޑ‬ਏ݀཮ᡉ๱ ‫ׯ‬๓(Skurichina & Duin, 1998*Ƕ !. ಃϤ࿯ Random Subspace Method ࣁΑှ،ӧଯᆢࡋΠǴ૽ግኬҁኧό‫ޑى‬௃‫܌ݩ‬೷ԋ‫ޑ‬ୢᚒǴҗ Ho (1998). 14.

(15) ‫܌‬ගр‫ ޑ‬RSM ࢂ΋ঁёаှ،λໆኬҁୢᚒ‫ޑ‬Б‫ݤ‬ǴΨࢂ΋ᅿёගଯϩᜪᏔਏ ૈϷᛙ‫ޑ܄ۓ‬ӭख़ϩᜪᏔ‫מ‬ѯǴᙖҗಔӝӭঁόӕ੝ቻ‫ޑ‬ϩᜪᏔٰቚуϩᜪၗૻ ‫ޑ‬ϕံ‫܄‬Ǵӆ೸ၸಔӝ‫ޑ‬฼ౣǴૈԖਏගଯਏૈ(Skurichina & Duin, 2001)ǶRSM ٬Ҕ‫ޑ‬ಃ΋‫؁‬ᡯࣁ೛‫ޑۓڰۓ‬η‫ޜ‬໔‫ޑ‬ᆢࡋ rǴΨ൩ࢂᒿᐒᒧ‫ ۓڰۓ‬r ঁᆢࡋ‫ޑ‬ ੝ቻ‫ޜ‬໔Ǵӆ೛‫܌ۓ‬ಔӝ‫ޑ‬ϩᜪᏔঁኧ BǶԜѦǴRSM ύǴࢂа֡΋ϩѲٰᒿᐒ ᒧ‫ڗ‬੝ቻǶਥᏵ(Ho, 1998)‫ ޑ‬RSM ‫ޑ‬ϐᄽᆉ‫ݤ‬Ǵ‫ۓځ‬ကӵΠǺ. [RSM ᄽᆉ‫]ݤ‬ Input: A data set D {( xi , y i ) | 1 d i d N }, xi  &  ƒ p , y i {1,  , L}  C , where yi is the label of xi, L is the number of classes, and N is the training sample size. A learning algorithm (base classifier) L A fixed subspace dimensionality r < p The number of the base learners B Output: Final hypothesis h final : & o C computed by the ensemble. BEGIN for i = 1 to B Db Subspace_ selection( D, r ) hb Learner( Db ). end h final ( x) arg max yC card ( B | hb ( x). y). END. ࣴ‫ز‬᛾ܴ(Skurichina & Duin, 2002)྽ኬҁኧКၗ਑ᆢࡋኧ࣬ჹλਔǴRSM ё аԖᡉ๱‫ׯ‬๓ਏ݀ǴЪ RSM ‫ڀ‬ഢӭख़ϩᜪᏔϷ੝ቻᒧ‫ڗ‬෧Ͽᆢࡋኧ‫ޑ‬ᓬᗺǴ‫܌‬а ੝ձ፾Ҕ‫ܭ‬λኬҁ‫ޑ‬௃‫׎‬Ƕ. 15.

(16) ಃΎ࿯ Weighted Random Subspace Method ਥᏵ(Kuo, Hsieh, Liu, & Chao, 2005*Ў᝘ύว౜ǴRSM ‫ޑ‬η૽ግኬҁ໣ࢂҗη‫ޜ‬ ໔‫܌‬ᒧ‫ڗ‬ᆢࡋٰ،‫ޑۓ‬Ǵՠ RSM ‫܌‬ᒧ‫ڗ‬ϐη‫ޜ‬໔ᆢࡋ r ࢂ‫ޑۓڰ‬ǴऩճҔਡѳ ྖϯ‫ٰݤ‬ჹ‫؂‬ԛ‫ ޑڗܜ܌‬r ຾Չ‫׳‬ཥǴ٠ࡌᄬᆢࡋ‫ޑ‬ख़ा‫܄‬ϩଛٰ຾Չ՗ीᆢࡋ ‫ޑ‬ϩଛǴஒаԜϩଛٰ‫ ڗܜ‬r ຾Զࡌᄬη૽ግኬҁ໣Ǵ೭ਔং‫ ޑ‬r ൩όࢂ‫ޑۓڰ‬ ΑǶW ࣁচԖ੝ቻ‫ޑ‬ख़ा‫܄‬ϩѲ‫ࣁځ‬ᚆණϩѲᜪࠠǴӚᜪࠠϐᄽᆉ‫ݤ‬ӵΠ‫܌‬ҢǺ ! [RSM-KS ᄽᆉ‫]ݤ‬ Input: A data set D {( xi , y i ) | 1 d i d N }, xi  &  ƒ p , y i {1,  , L}  C where yi is the label of xi, L is the number of classes, and N is the training sample size. A learning algorithm (base classifier) L The number of the base learners B The number of the base learners b0 Subspace importance distribution R An initial importance distribution of dimensionality R0 Uniform distribution U Output: Final hypothesis h final : & o C computed by the ensemble. BEGIN for k = 1 to b0 rk. 1  (k  1) * ¬min jC ( N j ) / b0 ¼. Dk. Subspace _ selection( D, rk ,U ). hk. Learner( Dk ). Compute and normalize ACC (hk ) as the initial distribution (R0) of R where ACC (hk ) is the re-substitution classification accuracy of training data by using hk end for i = b0 to B rb. Dimension _ selection( Rb1 ). Db. Subspace_ selection( D, rb ,U ). hb. Learner ( Db ). and Compute ACC (hb ). 16.

(17) Rb. Kernel _ smooth ( Rb 1 , rb , ACC ( hb )). end h final ( x). arg max tC card ( B | hb ( x). y). END! ! ‫ঁ؂‬ཥ૽ግኬҁ໣җ Subspace_selection ำ‫ׇ‬ёளǶӆ٩૽ግϩᜪᏔள‫ډ‬،฼ ϩᜪᏔ hbǴӆа hb ‫૽ޑ‬ግኬҁϩᜪ҅ዴ౗բࣁ៾ॶǴаਡѳྖϯ‫׳ٰݤ‬ཥ RǶԜ ၸำख़ᙟ B ԛǴӆճҔЬा‫׫‬౻‫ٰ่ݤ‬ӝϩᜪᏔள‫ډ‬നಖ،฼Ƕ߃‫ޑۈ‬ϩѲ R0 ࢂ җ b0 ঁ߃‫ۈ‬ᗺीᆉ‫ځ‬ᗺ‫ޑ‬ϩѲࢂவ1, ¬min iC ( N i ) / b0 ¼, , (b0  1)¬min iC ( N i ) / b0 ¼ ӧҁࣴ‫ز‬ ύǴb0 ࢂ೛‫ ࣁۓ‬5Ƕӧ߃‫ۈ‬ϯࡕǴᆢࡋኧ‫ޑ‬ख़ा‫܄‬ϩѲ R ࢂ࿶җਡѳྖϯ‫ڄ‬ኧٰ ‫׳‬ཥǴ‫ځ‬ϦԄ‫ۓ‬ကӵΠǺ f (r | ACC (hm ), m. B0 ,..., b). b. 1. ¦ ACC (h. m. )V. r  rm. ¦ ACC (h. )K (. 1. ( r  rm ) 2 ) 2V 2. m. p. m B0. V. ),. (2-18). i 1. K(. r  rm. V. ). 2SV 2. exp( . (2-19). ӧԜ V ࢂ஥ቨ(bandwidth)Ƕ [WRSM-KS1 ᄽᆉ‫]ݤ‬ Input: A data set D {( xi , y i ) | 1 d i d N }, xi  &  ƒ p , y i {1,  , L}  C where yi is the label of xi, L is the number of classes, and N is the training sample size. A learning algorithm (base classifier) L The number of the base learners B The number of the base learners b0 Subspace importance distribution R An initial importance distribution of dimensionality R0 Compute the accuracy of training data ACC An importance distribution of original features W Output: Final hypothesis h final : & o C computed by the ensemble. BEGIN. 17.

(18) for q = 1 to p hq. Learner({( xiq , yi ) | i 1,...N }). Wq. ACC (hq ). end for k = 1 to b0 rk. 1  (k  1) * ¬min jC ( N j ) / b0 ¼. Dk. Subspace _ selection( D, rk , W ). hk. Learner( Dk ). Compute and normalize ACC (hk ) as the initial distribution (R0) of R where ACC (hk ) is the re-substitution classification accuracy of training data by using hk end for i = b0 to B rb. Dimension _ selection( Rb1 ). Db. Subspace_ selection ( D, rb ,W ). hb. Learner ( Db ). Rb. Kernel _ smooth ( Rb 1 , rb , ACC ( hb )). and Compute ACC (hb ). end h final ( x). arg max tC card ( B | hb ( x). y). END WRSM-KS1 Б‫ׯ׳ݤ‬Α੝ቻ‫ޑ‬ᒧ‫ڗ‬БԄǴ‫ׯ‬җ૽ግኬҁঁձ‫ޑ‬ϩᜪ҅ ዴ౗ٰբࣁӚձ੝ቻ‫ޑ‬у៾Ǵ٠ࡌᄬ੝ቻख़ाำࡋϩଛ WǴ٠όᘐ‫׳‬ཥᆢ ࡋ‫ޑ‬ख़ा‫܄‬ϩѲ‫ڄ‬ኧ RǴԶௗΠٰ‫ޑ‬ᒧ‫ڗ‬੝ቻБ‫߾ݤ‬ᆶ RSM-KS ࣬ӕǴճ Ҕ‫׳‬ཥၸࡕख़ा‫܄‬ϩѲ‫ڄ‬ኧёкϩ‫ׯ‬຾ RSM а֡΋ϩଛ‫ڗܜ‬੝ቻ‫ޑ‬όӝ ౛‫܄‬Ǵӧჴሞᡍ᛾΢Ψ᛾ܴаԜБԄࡌᄬख़ा‫܄‬ϩѲ‫ڄ‬ኧёаၲ‫ډ‬ၨ٫‫ޑ‬ ϩᜪ҅ዴ౗(Kuo, Hsieh, Liu, & Chao, 2005*Ǵკ 2-2~2-4 ࣁӚϩᜪᏔ‫ࡌ܌‬ᄬ рٰ‫ ޑ‬R ϩѲǴёаவკ࣮р qdc ϩᜪᏔ‫ࡌ܌‬ᄬ‫ ޑ‬R ϩѲၨୃӛ‫ܭ‬եᆢ ࡋǴknnc ϩᜪᏔ‫ࡌ܌‬ᄬ‫ ޑ‬R ϩѲၨୃӛ‫ܭ‬ύ໔ᆢࡋǴsvc ϩᜪᏔ‫ࡌ܌‬ᄬ‫ޑ‬ R ϩѲၨୃӛ‫ܭ‬ၨଯ‫ޑ‬ᆢࡋǶ. 18.

(19) Density (Ø10-2 ). Density (Ø10-2 ). R Distribution (qdc). R Distribution (knnc) კ 2-3 WRSM-KS1, knnc ‫ ޑ‬R ϩѲ. Density (Ø10-2 ). კ 2-2 WRSM-KS1, qdc ‫ ޑ‬R ϩѲ. ! !. R Distribution (svc). !. კ 2-4 WRSM-KS1, svc ‫ ޑ‬R ϩѲ! ! [WRSM-KS2 ᄽᆉ‫]ݤ‬ Input: A data set D {( xi , y i ) | 1 d i d N }, xi  &  ƒ p , y i {1,  , L}  C where yi is the label of xi, L is the number of classes, and N is the training sample size. A learning algorithm (base classifier) L The number of the base learners B The number of the base learners b0 Subspace importance distribution R An initial importance distribution of dimensionality R0 Compute the accuracy of training data ACC An importance distribution of original features W Fisher’s linear discriminant analysis separability J Output: Final hypothesis h final : & o C computed by the ensemble. BEGIN for q = 1 to p Wq. J ({( xiq , yi ) | i 1,...N }). end for k = 1 to b0. 19.

(20) rk. 1  (k  1) * ¬min jC ( N j ) / b0 ¼. Dk. Subspace _ selection( D, rk , W ). hk. Learner( Dk ). Compute and normalize ACC (hk ) as the initial distribution (R0) of R where ACC (hk ) is the re-substitution classification accuracy of training data by using hk end for i = b0 to B rb. Dimension _ selection( Rb1 ). Db. Subspace_ selection ( D, rb ,W ). hb. Learner ( Db ). Rb. Kernel _ smooth ( Rb 1 , rb , ACC ( hb )). and Compute ACC (hb ). end h final ( x). arg max tC card ( B | hb ( x). y). END Fisher’s linear discriminant analysis separability is defined as follows: p. W _ LDAq. Jq / ¦ Ji , Jq. 1 trace( S wq S bq ) , q. 1,2..., p ! !. !. !. !. i 1. L. S wq. ¦P ¦ j. j 1. jq. , S bq. L. ¦ P (M j. jq.  M 0 q )( M. j 1. jq.  M 0 q )T , M 0q. L. ¦PM j. jq. (2-20). j 1. WRSM-KS2 Б‫ݤ‬Ψ΋ኬࢂ‫ׯ׳‬੝ቻ‫ޑ‬ᒧ‫ڗ‬БԄǴҗጕ‫୔܄‬ձϩ‫ޑ݋‬ന٫ϯ‫ڄ‬ ኧٰբࣁӚձ੝ቻ‫ޑ‬у៾Ǵ٠ࡌᄬ੝ቻख़ाำࡋϩଛ WǴԶᒧ‫ڗ‬੝ቻБ‫ݤ‬Ψᆶ RSM-KS ࣬ӕǴճҔԜ‫׳ٰݤ‬ཥख़ा‫܄‬ϩѲ‫ڄ‬ኧǴΨёၲ‫ډ‬ၨ٫‫ޑ‬ϩᜪ҅ዴ౗Ƕ! ӧ WRSM ύǴЬाࢂှ، RSM ‫ޑ‬ᆢࡋᒧ᏷ୢᚒǴҔਡѳྖϯ‫୏ٰݤ‬ᄊᒧ‫ڗ‬ η‫ޜ‬໔ᆢࡋǴӧϩᜪਔа੝ቻу៾Б‫ٰݤ‬Ծ୏ᒧ‫ڗ‬ᆢࡋǴ٬Ҕ૽ግኬҁϩᜪ҅ዴ ౗Ϸ LDA ‫ޑ‬ᜪಔϩᚆໆന٫ϯྗ߾ёаК RSM ٰ‫׳ޑ‬Ԗਏ౗ЪਏૈΨ཮Ԗܴᡉ ‫ׯ‬๓(Kuo, Hsieh, Liu, & Chao, 2005)Ƕ. 20.

(21) ಃΟക Bagging Ϸ WRSM ϐ่ӝ ҁകύϩࣁΟ࿯Ǵಃ΋࿯ࣁࣴ‫ۺཷز‬ǹಃΒ࿯ࣁᄽᆉ‫ݤ‬ǹಃΟ࿯ࣁၗ਑ඔॊǶ ၁ॊӵΠǺ! !. ಃ΋࿯ ࣴ‫ۺཷز‬ ӧҁ࿯྽ύஒ௖૸ࣁՖ٬Ҕ bagging ᆶ WRSM ϐ่ӝٰࡌᄬӭख़ϩᜪᏔǶӧ bagging ύҗ‫ܭ‬ѝჹ૽ግኬҁբᒿᐒख़ፄ‫؁ޑڗܜ‬ᡯǴ߾คԵቾჹ‫ܭ‬੝ቻᒧ᏷‫ޑ‬೽ ҽǴϼӭคҔЪ཮೷ԋᚇૻ‫ޑ‬੝ቻࢂόѸा‫ޑ‬ǴӢԜჹ‫ܭ‬੝ቻᒧ‫ޑڗ‬೽ҽሡा೏ Եቾ຾ٰωӝ౛Ƕ࣬ჹ‫ ޑ‬WRSM ѝჹ੝ቻ଺ᒧ‫ޑڗ‬೽ҽǴ٠คԵቾ૽ግኬҁኧ ‫ޑ‬ό‫ޑى‬ୢᚒǶ‫ࢂܭ‬ҁࣴ‫ز‬ஒᔆံ‫ٿ‬ᅿБ‫ঁݤ‬ձό‫ޑى‬ӦБǴ΋ԛ଺่ӝ‫୏ޑ‬ բǴౢғ΋ঁཥ‫ޑ‬ӭख़ϩᜪᏔ‫س‬಍Ǵа‫؃‬ှ،‫ٿ‬ᅿБ‫ঁݤ‬ձό‫ޑى‬ӦБǶ! !. ಃΒ࿯ ᄽᆉ‫ݤ‬ ҁࣴ‫ز‬ගрќ΋ᅿཥ‫ޑ‬ӭख़ϩᜪᏔϐ่ӝǴ२Ӄϟಏᄽᆉ‫ݤ‬ϣ৒Ǵჹၗ਑٬ Ҕ‫ٿ‬໘ࢤ‫ڗ‬ኬ‫؁‬ᡯǴ२Ӄ٬Ҕ bagging ‫מޑ‬ѯǴ຾Չჹ૽ግኬҁ‫ܜޑ‬ኬ٠ౢғཥ ‫૽ޑ‬ግၗ਑໣ǴԶཥ‫ޑ‬ၗ਑໣ӆϩձᔈҔ WRSM ‫؁‬ᡯǴ຾Չჹ੝ቻ‫ܜޑ‬ኬǴന ࡕӆ٬ҔЬा‫׫‬౻‫଺ݤ‬рനࡕ،฼ǴаΠࣁᄽᆉ‫ݤ‬ϐϟಏǶ!. ൘ǵʳ BG-RSM-KS ᄽᆉ‫ݤ‬ BG-RSM-KS Б‫ុۯࢂݤ‬চ RSM-KS ‫ݤ‬Ǵόၸ΋໒‫ۈ‬Ӄჹၗ਑໣բᒿᐒख़ፄ ‫ڗ‬ኬ‫୏ޑ‬բǴௗΠٰ‫؁‬ᡯᆶ RSM-KS ‫࣬ݤ‬ӕǶ!. 21.

(22) ! [BG-RSM-KS ᄽᆉ‫]ݤ‬ 1. Repeat for b 1,2,..., B ʳ (i)Take a bootstrap replicate X b of the training data set X (ii)RSM-KS procedure using training data set X b 2. Combine classifiers C b ( X ), b 1,2,..., B by majority voting (the most often predicted label) to a final decision rule E ( x) arg max ¦ G C y{1,..., L}. b. b. ( x ), y. where G i , j is the Kronecker symbol, and y 1,2,..., L is a decision (class label) of the classifier for two-class problem. !. ມǵʳ BG-WRSM-KS1 ᄽᆉ‫ݤ‬ BG-WRSM-KS1 Б‫ݤ‬Ψࢂ‫ុۯ‬চ WRSM-KS1 ‫ݤ‬Ǵ΋໒‫ۈ‬Ӄჹၗ਑໣բᒿᐒ ख़ፄ‫ڗ‬ኬ‫୏ޑ‬բǴௗΠٰ‫؁‬ᡯᆶ WRSM-KS1 ‫࣬ݤ‬ӕǶ! [BG-WRSM-KS1 ᄽᆉ‫]ݤ‬ 1. Repeat for b 1,2,..., B ʳ (i)Take a bootstrap replicate X b of the training data set X (ii)WRSM-KS1 procedure using training data set X b 2. Combine classifiers C b ( X ), b 1,2,..., B by majority voting (the most often predicted label) to a final decision rule E ( x) arg max ¦ G C y{1,..., L}. b. b. ( x ), y. where G i, j is the Kronecker symbol, and y 1,2,..., L is a decision (class label) of the classifier for two-class problem.. 22.

(23) ୖǵʳ BG-WRSM-KS2 ᄽᆉ‫ݤ‬ BG-WRSM-KS2 Б‫ݤ‬΋ኬࢂ‫ុۯ‬চ WRSM-KS2 ‫ݤ‬Ǵ΋໒‫ۈ‬Ӄჹၗ਑໣բᒿ ᐒख़ፄ‫ڗ‬ኬ‫୏ޑ‬բǴௗΠٰ‫؁‬ᡯᆶ WRSM-KS2 ‫࣬ݤ‬ӕǶ! [BG-WRSM-KS2 ᄽᆉ‫]ݤ‬ 1. Repeat for b 1,2,..., B ʳ (i)Take a bootstrap replicate X b of the training data set X (ii)WRSM-KS2 procedure using training data set X b 2. Combine classifiers C b ( X ), b 1,2,..., B by majority voting (the most often predicted label) to a final decision rule E ( x) arg max ¦ G C y{1,..., L}. b. b. ( x ), y. where G i , j is the Kronecker symbol, and y 1,2,..., L is a decision (class label) of the classifier for two-class problem.. ߄ 3-1 ҁࣴ‫܌ز‬٬Ҕᄽᆉ‫ݤ‬ӈԄ ӄӜ!. ᕭቪ!. ᇥܴ!. bagging and random. BG-RSM-KS. bagging ‫ ک‬RSM җਡѳྖϯ‫ݤ‬. subspace method based on. ‫܌‬՗ी‫ޑ‬η‫ޜ‬໔ᆢࡋϩѲ S ٰ. kernel smoothing. Ծ୏ᒧ‫ڗ‬η‫ޜ‬໔ᆢࡋ!. bagging and weighted. BG-WRSM-KS1. bagging ‫ک‬٬Ҕ૽ግኬҁϩᜪ. random subspace method. ҅ዴ౗բࣁ੝ቻख़ाำࡋϩѲ. based on kernel smoothing 1. X ‫ޑ‬у៾!. bagging and weighted. BG-WRSM-KS2. bagging ‫ک‬٬Ҕጕ‫୔܄‬ձϩ‫݋‬. random subspace method. ᜪձϩᚆໆբࣁ੝ቻख़ाำࡋ. based on kernel smoothing 2. ϩѲ X ‫ޑ‬у៾. 23.

(24) ಃΟ࿯ ၗ਑ඔॊ ӧҁക‫ޑ‬ჴᡍ೛ीύǴନΑाᡍ᛾ҁࣴ‫܌ز‬ගр‫ޑ‬ཥӭख़ϩᜪᏔࢂցૈКঁ ձ଺ bagging Ϸ WRSM ‫׳‬ԖਏૈǴΨ൩ࢂᇥૈցӧϩᜪ҅ዴ౗΢Ԗගϲ‫ޑ‬ਏ݀Ƕ ҁࣴ‫ز‬Ҟ߻٬ҔΑ‫ٿ‬ᅿόӕ‫ޑ‬ଯӀ᛼ᇿෳቹႽၗ਑໣аϷ௲‫ػ‬ෳᡍၗ਑଺ࣁჴ ᡍၗ਑Ƕ‫؂‬ᅿၗ਑ࣣԖ 10 ಔၗ਑໣ǴӚ໨ኧᏵࣁ 10 ಔၗ਑໣‫ޑ‬ѳ֡Ƕ. ൘ǵʳ Washington DC Mall ၗ਑໣ ٬Ҕ Washington DC Mall ೿ѱӦ୔‫ޑ‬ଯӀ᛼ቹႽၗ਑(Landgrebe, 2003)Ƕགෳ Ꮤவ 0.4 ‫ ډ‬2.4 µm ‫ ڗ‬210 ঁ‫ࢤݢ‬Ƕх֖ёຎӀ୔ୱϷϣआѦጕӀ᛼ǶӢࣁςӃ ѐନ௞Н‫܌‬೷ԋ‫ޑ‬ᚇૻǴࡺӧҁჴᡍѝ٬Ҕ 191 ‫ࢤݢ‬ǶӅԖΎঁᜪձϩձࢂࡌᑐ ‫(ނ‬roofs)ǵຉၰ(roads)ǵλၡ(paths)ǵ૛Ӧ(grass)ǵᐋ݅(trees)ǵН(water)Ϸ഍ቹ (shadow)வύᒧ᏷૽ግኬҁϷෳ၂ኬҁǴၗ਑໣྽ύ૽ግኬҁϩࣁ‫ঁ؂‬ᜪձӚ 20ǵ40ǵ100Ǵ٠‫ঁ؂ڗܜ‬ᜪձӚ 100 ঁෳ၂ኬҁǴϩࣁჴᡍ 1ǵჴᡍ 2ǵჴᡍ 3Ǵ ჴሞ‫ܡޜ‬კӵკ 3-1 ‫܌‬ҢǴԶၗ਑໣‫ޑ‬ჴᡍኬҁኧӵ߄ 3-2Ƕ! !. 24.

(25) კ 3-1 Washington DC Mall आѦጕ‫ܡޜ‬კ! ! ߄ 3-2 Washington DC Mall ၗ਑໣ჴᡍ೛ी ჴᡍ. ჴᡍ 1. ჴᡍ 2. ᜪձኧ. 7. ᆢࡋኧ. 191. ჴᡍ 3. ૽ግኬҁኧ (ঁձᜪձ) ᕴ૽ግኬҁ. 20. 40. 100. 140. 280. 700. ෳ၂ኬҁኧ 100. (ঁձᜪձ) ᕴෳ၂ኬҁ. 700. 25.

(26) ມǵʳ Indian Pine Site ၗ਑໣ ٬Ҕ Indian Pine Site AVIRIS ‫ޜ‬ၩଯӀ᛼ቹႽǴԜၗ਑ࣁ 1992 ԃ 6 ДӑಃӼ ٗԀՋч೽ 100 ѳБϦٚ୔ୱϐၭ཰ҔӦቹႽӵკΐǴ‫ڀ‬Ԗ 220 ঁԖਏᓎ᛼Ȑᆢ ࡋȑ ǴӅ 9 ঁᜪձϩձࢂҏԯҖςહӦ(corn-clean)ǵҏԯҖ҂હӦ(corn-notill)ǵҏ ԯౣહӦ(corn-min)ǵ‫ށ‬૛Ӧ(grass/pasture)ǵ݅Ӧ(woods)ǵଳ૛Ӧ(hay-windrowed)ǵ ε ‫ ل‬҂ હ Ӧ (soybean-notill) ǵ ε ‫ ل‬ౣ હ Ӧ (soybean-min) ‫ ک‬ε ‫ ل‬ρ હ Ӧ (soybean-clean*Ǵၗ਑໣྽ύ૽ግኬҁϩࣁ‫ঁ؂‬ᜪձӚ 20ǵ40ǵ100Ǵ٠‫ঁ؂ڗܜ‬ ᜪձӚ 200 ঁෳ၂ኬҁǴམଛ BG-WRSM Ӛᄽᆉ‫ݤ‬ϩࣁჴᡍ 4ǵჴᡍ 5ǵჴᡍ 6Ǵ ჴሞ‫ܡޜ‬კӵკ 3-2 ‫܌‬ҢǴၗ਑໣‫ޑ‬ჴᡍኬҁኧӵ߄ 3-3Ƕ! !. კ 3-2 Indian Pine Site आѦጕ‫ܡޜ‬კ!. 26.

(27) ߄ 3-3 Indian Pine Site ၗ਑໣ჴᡍ೛ी ჴᡍ. ჴᡍ 4. ჴᡍ 5. ᜪձኧ. 9. ᆢࡋኧ. 220. ჴᡍ 6. ૽ግኬҁኧ (ঁձᜪձ) ᕴ૽ግኬҁ. 20. 40. 100. 180. 360. 900. ෳ၂ኬҁኧ 200. (ঁձᜪձ) ᕴෳ၂ኬҁ. 1800. ୖǵʳ ௲‫ػ‬ෳᡍၗ਑໣ ҁࣴ‫ز‬٬Ҕ‫ޑ‬௲‫ػ‬ෳᡍၗ਑໣ IȐ೾դԽǴ҇ 95ȑࣁճҔȨՉࡹଣ୯ৎࣽᏢ‫ہ‬ ঩཮ံշ஑ᚒࣴ‫ز‬ीฝ-୯λኧᏢࣽႝတϯ፾‫܄‬ບᘐෳᡍȩಃ΋ԃࡼෳϐર฽ෳᡍၗ ਑ǴࡼෳൂϡࣁநଈЎ௲٣཰Ьጓϐ୯λኧᏢࣽಃΜ΋нಃΒൂϡȨᘉϩǵऊϩȩ Ƕ! ਥᏵȨᘉϩǵऊϩȩൂϡϐ௲‫׷‬ӦՏǴ࿶җኧՏ୯λ௲ৣᆶࣴ‫ز‬Γ঩૸ፕࡕǴ٩ ྣ‫ځ‬௲‫ޕ׷‬᛽่ᄬ຾Չံ௱௲Ꮲᜪࠠ‫ޑ‬ϩᜪǶҁൂϡર฽ෳᡍीԖ 27 ᚒǴԖਏኬҁ 1192 ঁҔа຾ՉჴᡍǴ૽ግኬҁϩࣁ‫ঁ؂‬ᜪձӚ 10 Ϸ 20Ǵམଛ BG-WRSM Ӛᄽᆉ ‫ݤ‬ϩࣁჴᡍ 7ǵჴᡍ 8Ǵᜪձၗ਑໣‫ޑ‬ჴᡍኬҁኧӵ߄ 3-4ǴਥᏵᒱᇤᜪࠠϩᜪᏢғ ಔձӵ߄ 3-5Ƕ!. ! ! ! 27.

(28) ߄ 3-4 ௲‫ػ‬ෳᡍၗ਑໣ I ჴᡍ೛ी ჴᡍ. ჴᡍ 7. ჴᡍ 8. ᜪձኧ. 15. ᆢࡋኧ. 27. ૽ግኬҁኧ 10. 20. ᕴ૽ግኬҁ. 150. 300. ᕴෳ၂ኬҁ. 1042. 892. (ঁձᜪձ). ! ! ߄ 3-5 ᘉϩǵऊϩൂϡᒱᇤᜪࠠϩಔ௃‫׎‬ ಔձ 1 2 3 4 5 6 7 8. Γኧ 89 31 186 154 62 41 80 59. 9. 63. 10. 59. 11. 79. 12. 77. 13. 35. 14 15 ӝी. 150 27 1192. ሡ຾Չံ௱௲Ꮲϐཷ‫ۺ‬ Ȩ‫ٿ‬౦ϩ҆Кၨελȩ! Ȩ‫ٿ‬౦ϩ҆Кၨελȩǵ Ȩ೯ϩȩ! Ȩനᙁϩኧȩ! Ȩനᙁϩኧȩǵ Ȩ‫ٿ‬౦ϩ҆Кၨελȩ! Ȩനᙁϩኧȩǵ Ȩ‫ٿ‬౦ϩ҆Кၨελȩǵ Ȩ೯ϩȩ! Ȩऊϩȩ! Ȩനᙁϩኧȩǵ Ȩऊϩȩǵ Ȩ‫ٿ‬౦ϩ҆Кၨελȩ! Ȩനᙁϩኧȩǵ Ȩऊϩȩǵ Ȩ‫ٿ‬౦ϩ҆Кၨελȩǵ Ȩ೯ϩȩ! Ȩനᙁϩኧȩǵ Ȩऊϩȩǵ ȨϦӢኧȩ ǵȨ฻ॶϩኧȩ ǵȨ‫ٿ‬౦ϩ ҆Кၨȩǵ Ȩ೯ϩȩ! ሡख़ཥᏢಞȨനᙁϩኧȩ ǵ Ȩऊϩȩ ǵ ȨϦӢኧȩ ǵ Ȩ฻ॶϩኧȩǵ Ȩ‫ٿ‬౦ϩ҆Кၨȩǵ Ȩ‫ٿ‬ӕϩ҆Кၨȩ ǵȨϦ७ኧȩ! Ȩനᙁϩኧȩǵ Ȩऊϩȩǵ Ȩ‫ٿ‬౦ϩ҆Кၨȩǵ Ȩ೯ϩȩǵ Ȩ‫ٿ‬ӕ ϩ҆Кၨ! Ȩനᙁϩኧȩ ǵ Ȩऊϩȩ ǵ Ȩ‫ٿ‬౦ϩ҆Кၨȩ ǵ Ȩ‫ٿ‬ӕϩ҆Кၨȩ ǵ ȨϦ७ኧȩ ǵȨᘉϩȩ! Ȩനᙁϩኧȩǵ Ȩऊϩȩǵ ȨϦӢኧȩ ǵȨ฻ॶϩኧȩ ǵȨ‫ٿ‬౦ϩ ҆Кၨȩǵ Ȩ‫ٿ‬ӕϩ҆Кၨȩ ǵȨϦ७ኧȩǵ Ȩᘉϩȩ! ‫܌‬аཷ‫ۺ‬೿ሡख़ཥᏢಞ! уமግಞȐಉЈҍᒱȑ! !. 28.

(29) Զ௲‫ػ‬ෳᡍၗ਑໣ IIȐ೾դԽǵֆች㧌ǵླྀਕ҇ǵ࢒ҥ଻ǵқৎᇬǴ҇ 92ȑࣁ நଈЎ௲٣཰Ьጓϐ୯λኧᏢࣽಃΜ΋нൂϡȨ৻‫׎‬ȩ Ƕҁൂϡર฽ෳᡍीԖ 21 ᚒǴ Ԗਏኬҁ 748 ঁҔа຾ՉჴᡍǴ૽ግኬҁϩࣁ‫ঁ؂‬ᜪձӚ 10 Ϸ 20Ǵམଛ BG-WRSM Ӛᄽᆉ‫ݤ‬ϩࣁჴᡍ 9ǵჴᡍ 10Ǵᜪձၗ਑໣‫ޑ‬ჴᡍኬҁኧӵ߄ 3-6ǴਥᏵᒱᇤᜪࠠϩ ᜪᏢғಔձӵ߄ 3-7Ƕ!. ! ߄ 3-6 ௲‫ػ‬ෳᡍၗ਑໣ II ჴᡍ೛ी ჴᡍ. ჴᡍ 9. ჴᡍ 10. ᜪձኧ. 8. ᆢࡋኧ. 21. ૽ግኬҁኧ 10. 20. ᕴ૽ግኬҁ. 80. 160. ᕴෳ၂ኬҁ. 668. 588. (ঁձᜪձ). ߄ 3-7 ৻ࠠൂϡᒱᇤᜪࠠϩಔ௃‫׎‬ ಔձ 1 2 3 4 5 6 7 8 ӝी. Γኧ 50 36 47 221 53 30 25 286 748. ሡ຾Չံ௱௲Ꮲϐཷ‫ۺ‬ уமግಞ)ಉЈҍᒱ*! Ȩፄӝ৻ࠠय़ᑈȩ! Ȩፄӝ৻ࠠय़ᑈȩǵ Ȩ୷ҁ৻ࠠय़ᑈȩ! Ȩ৻ࠠ‫ۓ‬ကȩǵ Ȩፄӝ৻ࠠय़ᑈȩǵ Ȩ୷ҁ৻ࠠय़ᑈȩ! Ȩკ‫׎‬ᛤᇙȩ! Ȩፄӝ৻ࠠय़ᑈȩǵ Ȩკ‫׎‬ᛤᇙȩ! Ȩፄӝ৻ࠠय़ᑈȩǵ Ȩ୷ҁ৻ࠠय़ᑈȩ ǵȨკ‫׎‬ᛤᇙȩ! ‫܌‬Ԗཷ‫ۺ‬೿໪ख़ཥᏢಞ! !. 29.

(30) ಃѤക!ࣴ‫่݀ز‬Ϸ૸ፕ! ௗΠٰҁക࿯ஒ૸ፕҁࣴ‫܌ز‬ගрཥБ‫ݤ‬ᆶᙑБ‫ݤ‬ϐჴᡍКၨ่݀ǴϩᜪᏔ ϩᜪϐ҅ዴ౗ӵ߄ 4-1~4-10 ‫܌‬ҢǴಉᡏ೽ҽࣁᏱԖന٫ϩᜪ҅ዴ౗‫ޑ‬ಔӝǶ! Washington!DC MALL ϩᜪ҅ዴ౗(߄ 5-1~5-3)Ǵӧλኬҁύ(ჴᡍ 1)ᜪձ૽ግ ኬҁࣁ 20 ਔǴൂ΋ϩᜪᏔ‫ޑ‬ന٫ϩᜪ҅ዴ౗ࣁ 0.838(knnc)Ǵ٬Ҕ WRSM ന٫ϩ ᜪ҅ዴ౗ࣁ 0.912(WRSM-KS2*Ǵ࿶җཥᄽᆉ‫ݤ‬ёගϲԿ 0.932(BG-WRSM-KS2* Ԗ 2%‫ޑ‬ගϲНྗǶӧ qdc Ϸ svc ϩᜪᏔ‫ޑ‬ཥБ‫ݤ‬ᆶᙑБ‫߄ݤ‬౜Бय़߾คϼεৡ౦Ǵ! კ 4-1 ࣁҁࣴ‫܌ز‬٬Ҕϐ Washington DC Mall ೽ϩआѦጕ‫ܡޜ‬კቹႽǴკ 4-5~4-7 ߾ࢂ‫ڗ‬ӧჴᡍ 1 ྽ύԖၨܴᡉৡຯϩᜪਏ݀‫ ޑ‬knnc ϩᜪᏔϐϩᜪ่݀კǴआ୮೽ ϩࣁϩᜪਏ݀Ԗৡ౦ϐೀǶ. კ 4-1 Washington DC Mall ೽ϩआѦጕ‫ܡޜ‬კቹႽ!. 30.

(31) ߄ 4-1 Washington DC Mall (Ni=20)ϩᜪ҅ዴ౗ Dmbttjgjfs! C!. Bmhpsjuin!. red! CH!. XSTN!. lood)l>2*! CH.!. CH!. XSTN!. XSTN!. twd! CH.!. CH!. XSTN!. XSTN!. 1/254!. CH.! XSTN!. 2!. TJOHMF!. 1/949!. 31!. LT!. 1/254!. 1/:34!. 1/:26!. 1/947!. 1/967!. 1/971!. 1/894!. 1/986!. 1/:21!. 31!. LT2!. 1/254!. 1/:46!. 1/:45!. 1/947!. 1/95:!. 1/967!. 1/894!. 1/:23!. 1/:37!. 31!. LT3!. 1/254!. 1/:41!. 1/:47!. 1/947!. 1/:18!. 1/:39!. 1/894!. 1/:45!. 1/:46!. 61!. LT!. 1/254!. 1/:42!. 1/:36!. 1/948!. 1/969!. 1/97:!. 1/8::!. 1/:14!. 1/:25!. 61!. LT2!. 1/254!. 1/:52!. 1/:47!. 1/948!. 1/968!. 1/978!. 1/8::!. 1/:38!. 1/:47!. 61!. LT3!. 1/254!. 1/:47!. 1/:49!. 1/948!. 1/:22!. 1/:43!. 1/8::!. 1/:46!. 1/:47!. 211!. LT!. 1/254!. 1/:43!. 1/:3:!. 1/949!. 1/96:!. 1/981!. 1/915!. 1/:18!. 1/:29!. 211!. LT2!. 1/254!. 1/:54!. 1/:48!. 1/949!. 1/967!. 1/979!. 1/915!. 1/:42!. 1/:46!. 211!. LT3!. 1/254!. 1/:48!. 1/:49!. 1/949!. 1/:23!. 1/:43!. 1/915!. 1/:53!. 1/:52!. 31. 1/673!.

(32) ϩᜪ҅ዴ౗. ˥̆̀ˀ˞˦. ˃ˁˌˈ. ˪̅̆̀ˀ˞˦˄. ˃ˁˌ. ˪̅̆̀ˀ˞˦˅. ˃ˁˋˈ. ˕˚ˀ˥̆̀ˀ˞˦. ! &. ˕˚ˀ˪̅̆̀ˀ˞˦˄. ˃ˁˋ ̄˷˶. ˾́́˶. ̆̉˶. ˕˚ˀ˪̅̆̀ˀ˞˦˅. ϩᜪᏔ კ 4-2 Washington DC Mall!ϩᜪ҅ዴ౗Кၨკ)ჴᡍ 1,Ni=20,B=20*! ϩᜪ҅ዴ౗. ˃ˁˌˈ. ˥̆̀ˀ˞˦. ˃ˁˌ. ˪̅̆̀ˀ˞˦˄. ˃ˁˋˈ. ˪̅̆̀ˀ˞˦˅ ˕˚ˀ˥̆̀ˀ˞˦. ˃ˁˋ. ! ! &. ˕˚ˀ˪̅̆̀ˀ˞˦˄. ˃ˁˊˈ ̄˷˶. ˾́́˶. ̆̉˶. ˕˚ˀ˪̅̆̀ˀ˞˦˅. ! ϩᜪᏔ კ 4-3 Washington DC Mall ϩᜪ҅ዴ౗Кၨკ)ჴᡍ 1,Ni=20,B=50*! ϩᜪ҅ዴ౗. ˃ˁˌˈ. ˥̆̀ˀ˞˦ ˪̅̆̀ˀ˞˦˄. ˃ˁˌ. ˪̅̆̀ˀ˞˦˅. ˃ˁˋˈ. ˕˚ˀ˥̆̀ˀ˞˦. ! ! &. ˕˚ˀ˪̅̆̀ˀ˞˦˄. ˃ˁˋ ̄˷˶. ˾́́˶. ̆̉˶. ˕˚ˀ˪̅̆̀ˀ˞˦˅. ϩᜪᏔ კ 4-4 Washington DC Mall ϩᜪ҅ዴ౗Кၨკ)ჴᡍ 1,Ni=20,B=100ȑ !. 32. !.

(33) კ 4-5 Washington DC Mall ٬Ҕ knnc ‫ޑ‬ϩᜪ่݀კ)Ni=20*. კ 4-6 Washington DC Mall ٬Ҕ knnc ‫ ک‬Wrsm-KS2 ‫ޑ‬ϩᜪ่݀კ)Ni=20,B=20*!. კ 4-7 Washington DC Mall ٬Ҕ knnc ‫ ک‬BG-Wrsm-KS2 ‫ޑ‬ϩᜪ่݀კ)Ni=20,B=20*!. 33.

(34) ӧჴᡍ 2 ύǴൂ΋ϩᜪᏔ‫ޑ‬ന٫ϩᜪ҅ዴ౗ࣁ 0.880(knnc)Ǵ٬Ҕ WRSM ന ٫ϩᜪ҅ዴ౗ࣁ 0.935(WRSM-KS2)࿶җཥᄽᆉ‫ݤ‬ёගϲԿ 0.952(BG-WRSM-KS2* Ԗ 1.7%‫ޑ‬ගϲНྗǴӧ qdc Ϸ svc ϩᜪᏔ‫ޑ‬ཥБ‫ݤ‬ᆶᙑБ‫߄ݤ‬౜Бय़߾คϼεৡ ౦Ǵόၸӧ svc ϩᜪᏔϐύǴཥᄽᆉ‫ݤ‬ᗋࢂёаၲ‫ډ‬ന٫ϩᜪ҅ዴ౗Ƕკ 4-11~4-13 ߾ࢂ‫ڗ‬ӧჴᡍ 2 ྽ύԖၨܴᡉৡຯϩᜪਏ݀‫ ޑ‬knnc ϩᜪᏔϐϩᜪ่݀კǴआ୮ ೽ϩࣁϩᜪਏ݀Ԗৡ౦ϐೀǶ! ! ߄ 4-2 Washington DC Mall (Ni=40)ϩᜪ҅ዴ౗! Dmbttjgjfs! C!. Bmhpsjuin!. red! CH!. XSTN!. lood)l>2*! CH.!. CH!. XSTN!. XSTN!. twd! CH.!. CH!. XSTN!. XSTN!. 1/254!. CH.! XSTN!. 2!. TJOHMF!. 1/991!. 31!. LT!. 1/254!. 1/:56!. 1/:45!. 1/98:!. 1/9:4!. 1/9:5!. 1/758!. 1/962!. 1/968!. 31!. LT2!. 1/254!. 1/:63!. 1/:55!. 1/98:!. 1/998!. 1/9:1!. 1/758!. 1/983!. 1/991!. 31!. LT3!. 1/254!. 1/:49!. 1/:48!. 1/98:!. 1/:45!. 1/:5:!. 1/758!. 1/9:3!. 1/9:8!. 61!. LT!. 1/254!. 1/:59!. 1/:49!. 1/992!. 1/9:8!. 1/9:4!. 1/799!. 1/963!. 1/964!. 61!. LT2!. 1/254!. 1/:66!. 1/:59!. 1/992!. 1/9:5!. 1/9:5!. 1/799!. 1/991!. 1/996!. 61!. LT3!. 1/254!. 1/:53!. 1/:51!. 1/992!. 1/:46!. 1/:63!. 1/799!. 1/9:3!. 1/9:6!. 211!. LT!. 1/254!. 1/:62!. 1/:47!. 1/991!. 1/9:8!. 1/9:9!. 1/7:9!. 1/976!. 1/979!. 211!. LT2!. 1/254!. 1/:68!. 1/:58!. 1/991!. 1/9:7!. 1/9:5!. 1/7:9!. 1/9:7!. 1/9:1!. 211!. LT3!. 1/254!. 1/:54!. 1/:4:!. 1/991!. 1/:45!. 1/:62!. 1/7:9!. 1/9:6!. 1/9::!. ! ! !. 34. 1/917!.

(35) ! ϩᜪ҅ዴ౗. ˄. ˥̆̀ˀ˞˦. ˃ˁˌˈ. ˪̅̆̀ˀ˞˦˄. ˃ˁˌ. ˪̅̆̀ˀ˞˦˅ ˕˚ˀ˥̆̀ˀ˞˦. ˃ˁˋˈ. ˕˚ˀ˪̅̆̀ˀ˞˦˄. ˃ˁˋ. ! ! &. ̄˷˶. !. ˾́́˶. ̆̉˶. ˕˚ˀ˪̅̆̀ˀ˞˦˅. ! ϩᜪᏔ კ 4-8 Washington DC Mall ϩᜪ҅ዴ౗Кၨკ)ჴᡍ 2,Ni=40,B=20*!. ϩᜪ҅ዴ౗. ˄. ˥̆̀ˀ˞˦. ˃ˁˌˈ. ˪̅̆̀ˀ˞˦˄. ˃ˁˌ. ˪̅̆̀ˀ˞˦˅ ˕˚ˀ˥̆̀ˀ˞˦. ˃ˁˋˈ. ! ! &. ˕˚ˀ˪̅̆̀ˀ˞˦˄. ˃ˁˋ ̄˷˶. ˾́́˶. ̆̉˶. ˕˚ˀ˪̅̆̀ˀ˞˦˅. ϩᜪᏔ კ 4-9 Washington DC Mall ϩᜪ҅ዴ౗Кၨკ(ჴᡍ 2,Ni=40,B=50) ϩᜪ҅ዴ౗. ˄. !. ˥̆̀ˀ˞˦. ˃ˁˌˈ. ˪̅̆̀ˀ˞˦˄. ˃ˁˌ. ˪̅̆̀ˀ˞˦˅ ˕˚ˀ˥̆̀ˀ˞˦. ˃ˁˋˈ. ! ! &. ˕˚ˀ˪̅̆̀ˀ˞˦˄. ˃ˁˋ ̄˷˶. ˾́́˶. ̆̉˶. ˕˚ˀ˪̅̆̀ˀ˞˦˅. ! ϩᜪᏔ კ 4-10 Washington DC Mall ϩᜪ҅ዴ౗Кၨკ(ჴᡍ 2,Ni=40,B=100) ! !. 35.

(36) კ 4-11 Washington DC Mall ٬Ҕ knnc ‫ޑ‬ϩᜪ่݀კ)Ni=40*. კ 4-12 Washington DC Mall ٬Ҕ knnc ‫ ک‬Wrsm-KS2 ‫ޑ‬ϩᜪ่݀კ)Ni=40,B=20*!. კ 4-13 Washington DC Mall ٬Ҕ knnc ‫ ک‬BG-Wrsm-KS2 ‫ޑ‬ϩᜪ่݀კ)Ni=40,B=20*. !. 36.

(37) ӧჴᡍ 3 ൂ΋ϩᜪᏔ‫ޑ‬ന٫ϩᜪ҅ዴ౗ࣁ 0.923(knnc)Ǵ٬Ҕ WRSM ന٫ϩ ᜪ҅ዴ౗ࣁ 0.953(WRSM-KS2*࿶җཥᄽᆉ‫ݤ‬ёගϲԿ 0.961(BG-WRSM-KS2*Ԗ 0.8%‫ޑ‬ගϲНྗǶӧ svc ϩᜪᏔϐύǴཥᄽᆉ‫ݤ‬Ψࢂёаၲ‫ډ‬ന٫ϩᜪ҅ዴ౗Ƕ ॶள‫ݙ‬ཀ‫ࢂޑ‬Ǵᕵ٬ӧ qdc ϩᜪᏔ‫ޑ‬ཥБ‫߾ݤ‬ค‫ݤ‬ၲ‫ډ‬ᙑБ‫߄ݤ‬౜НྗǴՠҔ knnc ϩᜪᏔΨёၲ‫ډ‬ᆶ qdc ϩᜪᏔ΋ኬଯ‫ޑ‬ϩᜪਏ݀Ǵ‫ࢂܭ‬ёа೭ኬᇥǴཥᄽᆉӧ Washington DC Mall ၗ਑໣྽ύǴknnc ϩᜪᏔᆶ svc ϩᜪᏔ‫ޑ‬ਏૈࢂКၨᛙ‫ۓ‬Ǵ ԶЪ knnc ϩᜪᏔࢂёаၲ‫ډ‬ᡉ๱‫ׯ‬๓‫ޑ‬ਏ݀Ǵკ 4-17~4-19 ߾ࢂ‫ڗ‬ӧჴᡍ 3 ྽ύ Ԗၨܴᡉৡຯϩᜪਏ݀‫ ޑ‬knnc ϩᜪᏔϐϩᜪ่݀კǴआ୮೽ϩࣁϩᜪਏ݀Ԗৡ ౦ϐೀǶ! ! ߄ 4-3 Washington DC Mall (Ni=100)ϩᜪ҅ዴ౗! Dmbttjgjfs! C!. Bmhpsjuin!. red! CH!. XSTN!. lood)l>2*! CH.!. CH!. XSTN!. XSTN!. twd! CH.!. CH!. XSTN!. XSTN!. 1/254!. CH.! XSTN!. 2!. TJOHMF!. 1/:34!. 31!. LT!. 1/437!. 1/:64!. 1/:2:!. 1/:33!. 1/:47!. 1/:47!. 1/:2:!. 1/:2:!. 1/:39!. 31!. LT2!. 1/437!. 1/:6:!. 1/:52!. 1/:33!. 1/:42!. 1/:45!. 1/:2:!. 1/:34!. 1/:3:!. 31!. LT3!. 1/437!. 1/:4:!. 1/:5:!. 1/:33!. 1/:63!. 1/:71!. 1/:2:!. 1/:38!. 1/:41!. 61!. LT!. 1/538!. 1/:68!. 1/:45!. 1/:33!. 1/:49!. 1/:49!. 1/:31!. 1/:36!. 1/:41!. 61!. LT2!. 1/538!. 1/:71!. 1/:5:!. 1/:33!. 1/:45!. 1/:47!. 1/:31!. 1/:37!. 1/:42!. 61!. LT3!. 1/538!. 1/:49!. 1/:5:!. 1/:33!. 1/:64!. 1/:6:!. 1/:31!. 1/:41!. 1/:42!. 211!. LT!. 1/582!. 1/:67!. 1/:44!. 1/:33!. 1/:51!. 1/:4:!. 1/:32!. 1/:36!. 1/:3:!. 211!. LT2!. 1/582!. 1/:72!. 1/:62!. 1/:33!. 1/:48!. 1/:49!. 1/:32!. 1/:37!. 1/:43!. 211!. LT3!. 1/582!. 1/:49!. 1/:59!. 1/:33!. 1/:64!. 1/:72!. 1/:32!. 1/:44!. 1/:45!. 37. 1/:31!.

(38) ! ϩᜪ҅ዴ౗. ˃ˁˌˋ ˃ˁˌˉ ˃ˁˌˇ ˃ˁˌ˅ ˃ˁˌ ˃ˁˋˋ. ˥̆̀ˀ˞˦ ˪̅̆̀ˀ˞˦˄ ˪̅̆̀ˀ˞˦˅ ˕˚ˀ˥̆̀ˀ˞˦. ! ! &. ˕˚ˀ˪̅̆̀ˀ˞˦˄ ̄˷˶. ˾́́˶. ̆̉˶. ˕˚ˀ˪̅̆̀ˀ˞˦˅. ! ϩᜪᏔ კ 4-14 Washington DC Mall ϩᜪ҅ዴ౗Кၨკ(ჴᡍ 3,Ni=100,B=20) ϩᜪ҅ዴ౗. ˃ˁˌˋ. ˥̆̀ˀ˞˦. ˃ˁˌˉ. ˪̅̆̀ˀ˞˦˄. ˃ˁˌˇ. ˪̅̆̀ˀ˞˦˅. ˃ˁˌ˅. ˕˚ˀ˥̆̀ˀ˞˦. ! ! &. ˕˚ˀ˪̅̆̀ˀ˞˦˄. ˃ˁˌ ̄˷˶. ˾́́˶. ̆̉˶. ˕˚ˀ˪̅̆̀ˀ˞˦˅. ! ϩᜪᏔ კ 4-15 Washington DC Mall ϩᜪ҅ዴ౗Кၨკ(ჴᡍ 3,Ni=100,B=50) ϩᜪ҅ዴ౗. ˃ˁˌˋ. ˥̆̀ˀ˞˦. ˃ˁˌˉ. ˪̅̆̀ˀ˞˦˄. ˃ˁˌˇ. ˪̅̆̀ˀ˞˦˅. ˃ˁˌ˅. ˕˚ˀ˥̆̀ˀ˞˦. ! ! &. ˕˚ˀ˪̅̆̀ˀ˞˦˄. ˃ˁˌ ̄˷˶. ˾́́˶. ̆̉˶. ˕˚ˀ˪̅̆̀ˀ˞˦˅. ! ϩᜪᏔ კ 4-16 Washington DC Mall ϩᜪ҅ዴ౗Кၨკ(ჴᡍ 3,Ni=100,B=100) . 38.

(39) კ 4-17 Washington DC Mall ٬Ҕ knnc ‫ޑ‬ϩᜪ่݀კ)Ni=100*. კ 4-18 Washington DC Mall ٬Ҕ knnc ‫ ک‬Wrsm-KS2 ‫ޑ‬ϩᜪ่݀კ(Ni=100,B=20). კ 4-19 Washington DC Mall ٬Ҕ knnc ‫ ک‬BG-Wrsm-KS2 ‫ޑ‬ϩᜪ่݀კ)Ni=100,B=20*. 39.

(40) ӧ Indian Pine Site ၗ਑໣(߄ 5-4~5-6)ύǴჴᡍ 4 ൂ΋ϩᜪᏔϩᜪਏૈᆶཥᄽ ᆉคܴᡉৡຯǴৡຯѝӧίϩՏ‫ޑ‬ᇤৡጄൎϐϣǴკ 4-20 ࣁ Indian Pine Site Ӧय़ ੿ჴ௃‫ݩ‬კǴკ 4-24~4-26 ࢂ‫ڗ‬ӧჴᡍ 4 ྽ύ qdc ϩᜪᏔϐϩᜪ่݀კǴԶკ 4-27~4-29 ߾ࢂ‫ڗ‬ӧჴᡍ 4 ྽ύ knnc ϩᜪᏔϐϩᜪ่݀კǴआ୮೽ϩࣁϩᜪਏ݀ Ԗৡ౦ϐೀǶ. კ 4-20 Indian Pine Site Ӧय़੿ჴ௃‫ݩ‬კ!. 40.

(41) ߄ 4-4 Indian Pine Site (Ni=20)ϩᜪ҅ዴ౗ Dmbttjgjfs! C!. Bmhpsjuin!. red! CH!. XSTN!. lood)l>2*! CH.!. CH!. XSTN!. XSTN!. twd! CH.!. CH!. XSTN!. XSTN!. 1/222!. CH.! XSTN!. 2!. TJOHMF!. 1/7:3!. 31!. LT!. 1/222!. 1/864!. 1/868!. 1/79:!. 1/7:4!. 1/796!. 1/886!. 1/828!. 1/818!. 31!. LT2!. 1/222!. 1/874!. 1/882!. 1/79:!. 1/7:5!. 1/797!. 1/886!. 1/82:!. 1/827!. 31!. LT3!. 1/222!. 1/828!. 1/835!. 1/79:!. 1/7:4!. 1/79:!. 1/886!. 1/82:!. 1/821!. 61!. LT!. 1/222!. 1/893!. 1/885!. 1/7:2!. 1/7:6!. 1/799!. 1/888!. 1/838!. 1/831!. 61!. LT2!. 1/222!. 1/8:4!. 1/885!. 1/7:2!. 1/7:5!. 1/7:2!. 1/888!. 1/84:!. 1/846!. 61!. LT3!. 1/222!. 1/863!. 1/848!. 1/7:2!. 1/7:6!. 1/7:8!. 1/888!. 1/83:!. 1/841!. 211!. LT!. 1/222!. 1/914!. 1/886!. 1/7:3!. 1/7:7!. 1/7:3!. 1/889!. 1/856!. 1/851!. 211!. LT2!. 1/222!. 1/923!. 1/888!. 1/7:3!. 1/7:5!. 1/7:2!. 1/889!. 1/857!. 1/852!. 211!. LT3!. 1/222!. 1/891!. 1/848!. 1/7:3!. 1/7:7!. 1/7::!. 1/889!. 1/853!. 1/849!. ! ! ! ! ! ! ! ! !. 41. 1/861!.

(42) ! ϩᜪ҅ዴ౗. ˃ˁˋ. ˥̆̀ˀ˞˦. ˃ˁˊˈ. ˪̅̆̀ˀ˞˦˄. ˃ˁˊ. ˪̅̆̀ˀ˞˦˅ ˕˚ˀ˥̆̀ˀ˞˦. ˃ˁˉˈ. ! &. ˕˚ˀ˪̅̆̀ˀ˞˦˄. ˃ˁˉ ̄˷˶. ˾́́˶. ̆̉˶. ˕˚ˀ˪̅̆̀ˀ˞˦˅. ϩᜪᏔ კ 4-21 Indian Pine Site ϩᜪ҅ዴ౗Кၨკ(ჴᡍ 4,Ni=20,B=20) ϩᜪ҅ዴ౗. ˃ˁˋˈ ˃ˁˋ ˃ˁˊˈ ˃ˁˊ ˃ˁˉˈ ˃ˁˉ. !. ˥̆̀ˀ˞˦ ˪̅̆̀ˀ˞˦˄ ˪̅̆̀ˀ˞˦˅ ˕˚ˀ˥̆̀ˀ˞˦. ! &. ˕˚ˀ˪̅̆̀ˀ˞˦˄ ̄˷˶. ˾́́˶. ̆̉˶. ˕˚ˀ˪̅̆̀ˀ˞˦˅. ϩᜪᏔ კ 4-22 Indian Pine Site ϩᜪ҅ዴ౗Кၨკ(ჴᡍ 4,Ni=20,B=50) ϩᜪ҅ዴ౗. ˃ˁˋˈ ˃ˁˋ ˃ˁˊˈ ˃ˁˊ ˃ˁˉˈ ˃ˁˉ. !. ˥̆̀ˀ˞˦ ˪̅̆̀ˀ˞˦˄ ˪̅̆̀ˀ˞˦˅ ˕˚ˀ˥̆̀ˀ˞˦. ! &. ˕˚ˀ˪̅̆̀ˀ˞˦˄ ̄˷˶. ˾́́˶. ̆̉˶. ˕˚ˀ˪̅̆̀ˀ˞˦˅. ϩᜪᏔ კ 4-23 Indian Pine Site ϩᜪ҅ዴ౗Кၨკ(ჴᡍ 4,Ni=20,B=100) ! ! !. 42. !.

(43) კ 4-24 Indian Pine Site ٬Ҕ qdc ‫ޑ‬ϩᜪ่݀კ)Ni=20*. კ 4-25 Indian Pine Site ٬Ҕ qdc ‫ ک‬Wrsm-KS1 ‫ޑ‬ϩᜪ่݀კ)Ni=20,B=20*!. კ 4-26 Indian Pine Site ٬Ҕ qdc ‫ ک‬BG-Wrsm-KS1 ‫ޑ‬ϩᜪ่݀კ)Ni=20,B=20*. 43.

(44) კ 4-27 Indian Pine Site ٬Ҕ knnc ‫ޑ‬ϩᜪ่݀კ)Ni=20*. კ 4-28 Indian Pine Site ٬Ҕ knnc ‫ ک‬Wrsm-KS2 ‫ޑ‬ϩᜪ่݀კ)Ni=20,B=20*. კ 4-29 Indian Pine Site ٬Ҕ knnc ‫ ک‬BG-Wrsm-KS2 ‫ޑ‬ϩᜪ่݀კ)Ni=20,B=20*. 44.

(45) ჴᡍ 5 ൂ΋ϩᜪᏔ‫ޑ‬ന٫ϩᜪ҅ዴ౗ࣁ 0.739(knnc)Ǵ٬Ҕ WRSM ന٫ϩᜪ ҅ዴ౗ࣁ 0.747(WRSM-KS2*࿶җ่ӝ฼ౣёගϲԿ 0.757(BG-WRSM-KS2*Ԗ 1%‫ޑ‬ගϲНྗǹ٬Ҕ qdc ϩᜪᏔਔЪӧϩᜪᏔঁኧࣁ 20 ‫ޑ‬ਔং WRSM ϩᜪ҅ ዴ౗ࣁ 0.833(WRSM-KS1*࿶җཥᄽᆉ‫ݤ‬ёගϲԿ 0.851(BG-WRSM-KS1*Ԗ 1.8% ‫ޑ‬ගϲǴԶӧ‫ځ‬дόӕঁኧϩᜪᏔБय़΋ኬёаၲ‫ډ‬ന٫ϩᜪ҅ዴ౗Ǵӧ svc ϩ ᜪᏔύϩᜪਏ݀εठ࣬ӕǶკ 4-33~4-35 ࢂ‫ڗ‬ӧჴᡍ 5 ྽ύԖၨܴᡉৡຯϩᜪਏ ݀‫ ޑ‬qdc ϩᜪᏔϐϩᜪ่݀კǴკ 4-36~4-38 ߾ࢂ knnc ϩᜪᏔϐϩᜪ่݀კǴआ ୮೽ϩࣁϩᜪਏ݀Ԗৡ౦ϐೀǶ ! ߄ 4-5 Indian Pine Site (Ni=40)ϩᜪ҅ዴ౗! Dmbttjgjfs! C!. Bmhpsjuin!. red! CH!. XSTN!. lood)l>2*! CH.!. CH!. XSTN!. XSTN!. twd! CH.!. CH!. XSTN!. XSTN!. 1/222!. CH.! XSTN!. 2!. TJOHMF!. 1/84:!. 31!. LT!. 1/222!. 1/935!. 1/957!. 1/848!. 1/852!. 1/859!. 1/8:1!. 1/83:!. 1/845!. 31!. LT2!. 1/222!. 1/944!. 1/962!. 1/848!. 1/852!. 1/856!. 1/8:1!. 1/857!. 1/852!. 31!. LT3!. 1/222!. 1/916!. 1/943!. 1/848!. 1/858!. 1/866!. 1/8:1!. 1/817!. 1/822!. 61!. LT!. 1/222!. 1/967!. 1/973!. 1/84:!. 1/854!. 1/864!. 1/8:4!. 1/848!. 1/84:!. 61!. LT2!. 1/222!. 1/971!. 1/975!. 1/84:!. 1/854!. 1/85:!. 1/8:4!. 1/856!. 1/854!. 61!. LT3!. 1/222!. 1/949!. 1/961!. 1/84:!. 1/857!. 1/867!. 1/8:4!. 1/816!. 1/826!. 211!. LT!. 1/222!. 1/977!. 1/977!. 1/84:!. 1/854!. 1/864!. 1/8:1!. 1/857!. 1/861!. 211!. LT2!. 1/222!. 1/979!. 1/97:!. 1/84:!. 1/853!. 1/861!. 1/8:1!. 1/862!. 1/862!. 211!. LT3!. 1/222!. 1/95:!. 1/965!. 1/84:!. 1/856!. 1/868!. 1/8:1!. 1/843!. 1/84:!. !. 45. 1/749!.

(46) ! ϩᜪ҅ዴ౗. ˄ ˃ˁˋ ˃ˁˉ ˃ˁˇ ˃ˁ˅ ˃. ˥̆̀ˀ˞˦ ˪̅̆̀ˀ˞˦˄ ˪̅̆̀ˀ˞˦˅ ˕˚ˀ˥̆̀ˀ˞˦. ! &. ˕˚ˀ˪̅̆̀ˀ˞˦˄ ̄˷˶. ˾́́˶. ̆̉˶. ˕˚ˀ˪̅̆̀ˀ˞˦˅. ϩᜪᏔ კ 4-30 Indian Pine Site ϩᜪ҅ዴ౗Кၨკ(ჴᡍ 5,Ni=40,B=20) ϩᜪ҅ዴ౗. ˄ ˃ˁˋ ˃ˁˉ ˃ˁˇ ˃ˁ˅ ˃. !. ˥̆̀ˀ˞˦ ˪̅̆̀ˀ˞˦˄ ˪̅̆̀ˀ˞˦˅ ˕˚ˀ˥̆̀ˀ˞˦. ! &. ˕˚ˀ˪̅̆̀ˀ˞˦˄ ̄˷˶. ˾́́˶. ̆̉˶. ˕˚ˀ˪̅̆̀ˀ˞˦˅. ϩᜪᏔ კ 4-31 Indian Pine Site ϩᜪ҅ዴ౗Кၨკ(ჴᡍ 5,Ni=40,B=50) ϩᜪ҅ዴ౗. ˃ˁˌ ˃ˁˋˈ ˃ˁˋ ˃ˁˊˈ ˃ˁˊ ˃ˁˉˈ. !. ˥̆̀ˀ˞˦ ˪̅̆̀ˀ˞˦˄ ˪̅̆̀ˀ˞˦˅ ˕˚ˀ˥̆̀ˀ˞˦. ! &. ˕˚ˀ˪̅̆̀ˀ˞˦˄ ̄˷˶. ˾́́˶. ̆̉˶. ˕˚ˀ˪̅̆̀ˀ˞˦˅. ϩᜪᏔ კ 4-32 Indian Pine Site ϩᜪ҅ዴ౗Кၨკ(ჴᡍ 5,Ni=40,B=100) ! ! !. 46. !.

(47) კ 4-33 Indian Pine Site ٬Ҕ qdc ‫ޑ‬ϩᜪ่݀კ)Ni=40*. კ 4-34 Indian Pine Site ٬Ҕ qdc ‫ ک‬Wrsm-KS1 ‫ޑ‬ϩᜪ่݀კ)Ni=40,B=20*. კ 4-35 Indian Pine Site ٬Ҕ qdc ‫ ک‬BG-Wrsm-KS1 ‫ޑ‬ϩᜪ่݀კ)Ni=40,B=20*. 47.

(48) კ 4-36 Indian Pine Site ٬Ҕ knnc ‫ޑ‬ϩᜪ่݀კ)Ni=40*. კ 4-37 Indian Pine Site ٬Ҕ knnc ‫ ک‬Wrsm-KS2 ‫ޑ‬ϩᜪ่݀კ)Ni=40,B=20*. კ 4-38 Indian Pine Site ٬Ҕ knnc ‫ ک‬BG-Wrsm-KS2 ‫ޑ‬ϩᜪ่݀კ)Ni=40,B=20*. 48.

(49) ჴᡍ 6 ൂ΋ϩᜪᏔ‫ޑ‬ന٫ϩᜪ҅ዴ౗ࣁ 0.794(knnc)Ǵൂᐱ٬Ҕ WRSM ന٫ ϩᜪ҅ዴ౗ࣁ 0.807(WRSM-KS2*࿶җ่ӝ฼ౣёගϲԿ 0.816(BG-WRSM-KS2*! Ԗ 0.9%‫ޑ‬ගϲНྗǴ٬Ҕ qdc ϩᜪᏔਔ٠ЪӧϩᜪᏔঁኧࣁ 20 ‫ޑ‬ਔং WRSM ϩ ᜪ҅ዴ౗ࣁ 0.874(WRSM-KS1*࿶җ่ӝ฼ౣёගϲԿ 0.893(BG-WRSM-KS1*Ԗ 1.9%‫ޑ‬ගϲǴԶӧ‫ځ‬дঁኧ‫ޑ‬ϩᜪᏔ΋ኬૈၲ‫ډ‬ϩᜪന٫ॶǴΨว౜ӧ Indian Pine Site ၗ਑໣྽ύ٬Ҕ(BG-WRSM-KS2*ᄽᆉ‫ૈݤ‬ၲ‫ډ‬ϩᜪന٫ॶ 0.904Ƕӧ svc ϩ ᜪᏔύϩᜪਏ݀εठ࣬ӕǴკ 4-42~4-44 ࢂ‫ڗ‬ӧჴᡍ 6 ྽ύԖၨܴᡉৡຯϩᜪਏ ݀‫ ޑ‬qdc ϩᜪᏔϐϩᜪ่݀კǴკ 4-45~4-47 ߾ࢂ knnc ϩᜪᏔϐϩᜪ่݀კǴआ ୮೽ϩࣁϩᜪਏ݀Ԗৡ౦ϐೀǶ! ! ߄ 4-6 Indian Pine Site (Ni=100)ϩᜪ҅ዴ౗! Dmbttjgjfs! C!. Bmhpsjuin!. red! CH!. XSTN!. lood)l>2*! CH.!. CH!. XSTN!. XSTN!. twd! CH.!. CH!. XSTN!. XSTN!. 1/222!. CH.! XSTN!. 2!. TJOHMF!. 1/8:5!. 31!. LT!. 1/222!. 1/981!. 1/992!. 1/8:3!. 1/8::!. 1/918!. 1/893!. 1/952!. 1/948!. 31!. LT2!. 1/222!. 1/985!. 1/9:4!. 1/8:3!. 1/912!. 1/8:3!. 1/893!. 1/963!. 1/962!. 31!. LT3!. 1/222!. 1/961!. 1/979!. 1/8:3!. 1/914!. 1/922!. 1/893!. 1/961!. 1/955!. 61!. LT!. 1/222!. 1/999!. 1/9::!. 1/8:5!. 1/913!. 1/91:!. 1/89:!. 1/964!. 1/957!. 61!. LT2!. 1/222!. 1/9:2!. 1/:13!. 1/8:5!. 1/912!. 1/916!. 1/89:!. 1/969!. 1/963!. 61!. LT3!. 1/222!. 1/976!. 1/994!. 1/8:5!. 1/917!. 1/926!. 1/89:!. 1/966!. 1/953!. 211!. LT!. 1/222!. 1/9:8!. 1/:13!. 1/8:5!. 1/913!. 1/921!. 1/8:4!. 1/972!. 1/96:!. 211!. LT2!. 1/222!. 1/9:7!. 1/:15!. 1/8:5!. 1/912!. 1/915!. 1/8:4!. 1/977!. 1/977!. 211!. LT3!. 1/222!. 1/983!. 1/99:!. 1/8:5!. 1/918!. 1/927!. 1/8:4!. 1/975!. 1/969!. 49. 1/899!.

(50) ! ϩᜪ҅ዴ౗. ˃ˁˌˈ ˃ˁˌ ˃ˁˋˈ ˃ˁˋ ˃ˁˊˈ ˃ˁˊ. ˥̆̀ˀ˞˦ ˪̅̆̀ˀ˞˦˄ ˪̅̆̀ˀ˞˦˅ ˕˚ˀ˥̆̀ˀ˞˦. ! &. ˕˚ˀ˪̅̆̀ˀ˞˦˄ ̄˷˶. ˾́́˶. ̆̉˶. ˕˚ˀ˪̅̆̀ˀ˞˦˅. ϩᜪᏔ კ 4-39 Indian Pine Site ϩᜪ҅ዴ౗Кၨკ(ჴᡍ 6,Ni=100,B=20) ϩᜪ҅ዴ౗. ˃ˁˌˈ. !. ˥̆̀ˀ˞˦. ˃ˁˌ. ˪̅̆̀ˀ˞˦˄. ˃ˁˋˈ. ˪̅̆̀ˀ˞˦˅ ˕˚ˀ˥̆̀ˀ˞˦. ˃ˁˋ. ! &. ˕˚ˀ˪̅̆̀ˀ˞˦˄. ˃ˁˊˈ ̄˷˶. ˾́́˶. ̆̉˶. ˕˚ˀ˪̅̆̀ˀ˞˦˅. ! კ 4-40 Indian Pine Site ϩᜪ҅ዴ౗Кၨკ(ჴᡍ 6,Ni=100,B=50) ϩᜪ҅ዴ౗. ˃ˁˌˈ ˃ˁˌ ˃ˁˋˈ ˃ˁˋ ˃ˁˊˈ ˃ˁˊ. !. ˥̆̀ˀ˞˦ ˪̅̆̀ˀ˞˦˄ ˪̅̆̀ˀ˞˦˅ ˕˚ˀ˥̆̀ˀ˞˦. ! &. ˕˚ˀ˪̅̆̀ˀ˞˦˄ ̄˷˶. ˾́́˶. ̆̉˶. ˕˚ˀ˪̅̆̀ˀ˞˦˅. ϩᜪᏔ კ 4-41 Indian Pine Site ϩᜪ҅ዴ౗Кၨკ(ჴᡍ 6,Ni=100,B=100) ! ! !. 50. !.

(51) ! კ 4-42 Indian Pine Site ٬Ҕ qdc ‫ޑ‬ϩᜪ่݀კ)Ni=100*. კ 4-43 Indian Pine Site ٬Ҕ qdc ‫ ک‬Wrsm-KS1 ‫ޑ‬ϩᜪ่݀კ)Ni=100,B=20*. კ 4-44 Indian Pine Site ٬Ҕ qdc ‫ ک‬BG-Wrsm-KS1 ‫ޑ‬ϩᜪ่݀კ)Ni=100,B=20*. 51.

(52) კ 4-45 Indian Pine Site ٬Ҕ knnc ‫ޑ‬ϩᜪ่݀კ)Ni=100*. კ 4-46 Indian Pine Site ٬Ҕ knnc ‫ ک‬Wrsm-KS2 ‫ޑ‬ϩᜪ่݀კ)Ni=100,B=20*. კ 4-47 Indian Pine Site ٬Ҕ knnc ‫ ک‬BG-Wrsm-KS2 ‫ޑ‬ϩᜪ่݀კ)Ni=100,B=20*!. 52.

(53) ӧ௲‫ػ‬ෳᡍၗ਑໣ I ύǴε೽ҽϩᜪᏔΠǴ٬Ҕӭख़ϩᜪᏔ‫ޑ‬ϩᜪਏ݀٠ؒ ԖКൂ΋ϩᜪᏔ‫ޑ‬ϩᜪਏٰ݀‫ޑ‬ӳǶԶЪว౜‫ډ‬Ҕ svc ൂ΋ϩᜪᏔ‫܌‬ၲ‫ޑډ‬ਏૈ ཮К‫ځ‬дϩᜪᏔٰ‫ޑ‬ӳࡐӭǴаΠࣁӚϩᜪᏔϐϩᜪ่݀Ǻ ! ߄ 4-7 ௲‫ػ‬ෳᡍၗ਑໣ I (Ni=10)ϩᜪ҅ዴ౗ Dmbttjgjfs! C!. Bmhpsjuin!. red! CH!. XSTN!. lood)l>2*! CH.!. CH!. XSTN!. XSTN!. twd! CH.!. CH!. XSTN!. XSTN!. 1/187!. CH.! XSTN!. 2!. TJOHMF!. 1/592!. 31!. LT!. 1/187!. 1/1:5!. 1/187!. 1/58:!. 1/257!. 1/1:6!. 1/719!. 1/3:8!. 1/377!. 31!. LT2!. 1/187!. 1/1:9!. 1/187!. 1/58:!. 1/264!. 1/212!. 1/719!. 1/45:!. 1/3::!. 31!. LT3!. 1/187!. 1/18:!. 1/187!. 1/58:!. 1/2:2!. 1/212!. 1/719!. 1/4:1!. 1/436!. 61!. LT!. 1/195!. 1/1:2!. 1/187!. 1/591!. 1/251!. 1/1:6!. 1/725!. 1/43:!. 1/412!. 61!. LT2!. 1/195!. 1/1:6!. 1/187!. 1/591!. 1/273!. 1/1:9!. 1/725!. 1/491!. 1/449!. 61!. LT3!. 1/195!. 1/18:!. 1/187!. 1/591!. 1/292!. 1/211!. 1/725!. 1/538!. 1/49:!. 211!. LT!. 1/196!. 1/1:7!. 1/187!. 1/593!. 1/253!. 1/1:9!. 1/733!. 1/492!. 1/457!. 211!. LT2!. 1/196!. 1/212!. 1/187!. 1/593!. 1/264!. 1/213!. 1/733!. 1/539!. 1/496!. 211!. LT3!. 1/196!. 1/17:!. 1/187!. 1/593!. 1/2:1!. 1/215!. 1/733!. 1/545!. 1/4:9!. ! ! ! ! ! !. 53. 1/778!.

(54) ! ϩᜪ҅ዴ౗. ˃ˁˈ ˃ˁˇ ˃ˁˆ ˃ˁ˅ ˃ˁ˄ ˃. ˥̆̀ˀ˞˦ ˪̅̆̀ˀ˞˦˄ ˪̅̆̀ˀ˞˦˅ ˕˚ˀ˥̆̀ˀ˞˦. ! &. ˕˚ˀ˪̅̆̀ˀ˞˦˄ ̄˷˶. ˾́́˶. ̆̉˶. ˕˚ˀ˪̅̆̀ˀ˞˦˅. ! ϩᜪᏔ კ 4-48 ௲‫ػ‬ෳᡍၗ਑໣ϩᜪ҅ዴ౗Кၨკ(ჴᡍ 7, Ni=10,B=20) ϩᜪ҅ዴ౗. ˃ˁˈ ˃ˁˇ ˃ˁˆ ˃ˁ˅ ˃ˁ˄ ˃. ˥̆̀ˀ˞˦ ˪̅̆̀ˀ˞˦˄ ˪̅̆̀ˀ˞˦˅ ˕˚ˀ˥̆̀ˀ˞˦. ! &. ˕˚ˀ˪̅̆̀ˀ˞˦˄ ̄˷˶. ˾́́˶. ̆̉˶. ˕˚ˀ˪̅̆̀ˀ˞˦˅. ! ϩᜪᏔ კ 4-49 ௲‫ػ‬ෳᡍၗ਑໣ϩᜪ҅ዴ౗Кၨკ(ჴᡍ 7, Ni=10,B=50) ϩᜪ҅ዴ౗. ˃ˁˈ ˃ˁˇ ˃ˁˆ ˃ˁ˅ ˃ˁ˄ ˃. ˥̆̀ˀ˞˦ ˪̅̆̀ˀ˞˦˄ ˪̅̆̀ˀ˞˦˅ ˕˚ˀ˥̆̀ˀ˞˦. ! &. ˕˚ˀ˪̅̆̀ˀ˞˦˄ ̄˷˶. ˾́́˶. ̆̉˶. ˕˚ˀ˪̅̆̀ˀ˞˦˅. ! ϩᜪᏔ კ 4-50 ௲‫ػ‬ෳᡍၗ਑໣ϩᜪ҅ዴ౗Кၨკ(ჴᡍ 7, Ni=10,B=100) ! ! !. 54.

(55) ! ߄ 4-8 ௲‫ػ‬ෳᡍၗ਑໣ I (Ni=20)ϩᜪ҅ዴ౗ Dmbttjgjfs! C!. Bmhpsjuin!. red! CH!. XSTN!. lood)l>2*! CH.!. CH!. XSTN!. XSTN!. twd! CH.!. CH!. XSTN!. XSTN!. 1/188!. CH.! XSTN!. 2!. TJOHMF!. 1/639!. 31!. LT!. 1/15:!. 1/227!. 1/188!. 1/627!. 1/385!. 1/238!. 1/761!. 1/553!. 1/445!. 31!. LT2!. 1/15:!. 1/211!. 1/188!. 1/627!. 1/432!. 1/243!. 1/761!. 1/627!. 1/512!. 31!. LT3!. 1/15:!. 1/225!. 1/188!. 1/627!. 1/454!. 1/272!. 1/761!. 1/631!. 1/534!. 61!. LT!. 1/159!. 1/231!. 1/188!. 1/636!. 1/3:6!. 1/251!. 1/75:!. 1/5:3!. 1/4:6!. 61!. LT2!. 1/159!. 1/223!. 1/188!. 1/636!. 1/474!. 1/258!. 1/75:!. 1/639!. 1/544!. 61!. LT3!. 1/159!. 1/239!. 1/188!. 1/636!. 1/482!. 1/299!. 1/75:!. 1/651!. 1/557!. 211!. LT!. 1/15:!. 1/22:!. 1/188!. 1/638!. 1/41:!. 1/261!. 1/764!. 1/665!. 1/576!. 211!. LT2!. 1/15:!. 1/1:8!. 1/188!. 1/638!. 1/4:1!. 1/268!. 1/764!. 1/697!. 1/5:3!. 211!. LT3!. 1/15:!. 1/221!. 1/188!. 1/638!. 1/48:!. 1/2::!. 1/764!. 1/693!. 1/612!. ! ! ! ! ! ! ! ! !. 55. 1/891!.

(56) ! ϩᜪ҅ዴ౗. ˃ˁˉ. ˥̆̀ˀ˞˦. ˃ˁˇ. ˪̅̆̀ˀ˞˦˄ ˪̅̆̀ˀ˞˦˅. ˃ˁ˅. ˕˚ˀ˥̆̀ˀ˞˦ ˕˚ˀ˪̅̆̀ˀ˞˦˄. ˃. ! &. ̄˷˶. ˾́́˶. ̆̉˶. ˕˚ˀ˪̅̆̀ˀ˞˦˅. ! ϩᜪᏔ კ 4-51 ௲‫ػ‬ෳᡍၗ਑໣ϩᜪ҅ዴ౗Кၨკ(ჴᡍ 8, Ni=20,B=20) ϩᜪ҅ዴ౗. ˃ˁˉ. ˥̆̀ˀ˞˦. ˃ˁˇ. ˪̅̆̀ˀ˞˦˄ ˪̅̆̀ˀ˞˦˅. ˃ˁ˅. ˕˚ˀ˥̆̀ˀ˞˦ ˕˚ˀ˪̅̆̀ˀ˞˦˄. ˃. ! &. ̄˷˶. ˾́́˶. ̆̉˶. ˕˚ˀ˪̅̆̀ˀ˞˦˅. ! ϩᜪᏔ კ 4-52 ௲‫ػ‬ෳᡍၗ਑໣ϩᜪ҅ዴ౗Кၨკ(ჴᡍ 8, Ni=20,B=50) ϩᜪ҅ዴ౗. ˃ˁˋ. ˥̆̀ˀ˞˦. ˃ˁˉ. ˪̅̆̀ˀ˞˦˄. ˃ˁˇ. ˪̅̆̀ˀ˞˦˅. ˃ˁ˅. ˕˚ˀ˥̆̀ˀ˞˦ ˕˚ˀ˪̅̆̀ˀ˞˦˄. ˃. ! &. ̄˷˶. ˾́́˶. ̆̉˶. ˕˚ˀ˪̅̆̀ˀ˞˦˅. ! ϩᜪᏔ კ 4-53 ௲‫ػ‬ෳᡍၗ਑໣ϩᜪ҅ዴ౗Кၨკ(ჴᡍ 8, Ni=20,B=100) ! ! !. 56.

(57) ӧ௲‫ػ‬ෳᡍၗ਑໣ II ύǴёаว౜‫܌ډ‬ԖϩᜪᏔ่݀ύǴ٬Ҕӭख़ϩᜪᏔ‫ޑ‬ ϩᜪਏ݀٠ؒԖКൂ΋ϩᜪᏔ‫ޑ‬ϩᜪਏٰ݀‫ޑ‬ӳǶ‫ࢂܭ‬ёаᇥҁࣴ‫܌ز‬ගрϐཥ ‫ޑ‬ӭख़ϩᜪᏔ‫س‬಍ӧ௲‫ػ‬ෳᡍၗ਑໣ύค‫ݤ‬Ԗ‫ׯ‬๓ਏૈ‫ޜޑ‬໔ǴԶ΋ኬว౜‫ډ‬Ҕ svc ϩᜪᏔ‫܌‬ၲ‫ޑډ‬ਏૈ཮К‫ځ‬дϩᜪᏔٰ‫ޑ‬ӳࡐӭǴаΠࣁӚϩᜪᏔϐϩᜪ่ ݀Ǻ ! ߄ 4-9 ௲‫ػ‬ෳᡍၗ਑໣ II (Ni=10)ϩᜪ҅ዴ౗ Dmbttjgjfs! C!. Bmhpsjuin!. red! CH!. XSTN!. lood)l>2*! CH.!. CH!. XSTN!. XSTN!. twd! CH.!. CH!. XSTN!. XSTN!. 1/454!. CH.! XSTN!. 2!. TJOHMF!. 1/436!. 31!. LT!. 1/348!. 1/171!. 1/171!. 1/426!. 1/243!. 1/226!. 1/635!. 1/55:!. 1/469!. 31!. LT2!. 1/348!. 1/197!. 1/171!. 1/426!. 1/254!. 1/235!. 1/635!. 1/571!. 1/473!. 31!. LT3!. 1/348!. 1/171!. 1/171!. 1/426!. 1/341!. 1/322!. 1/635!. 1/579!. 1/482!. 61!. LT!. 1/381!. 1/171!. 1/171!. 1/433!. 1/264!. 1/248!. 1/633!. 1/636!. 1/524!. 61!. LT2!. 1/381!. 1/171!. 1/171!. 1/433!. 1/264!. 1/252!. 1/633!. 1/628!. 1/533!. 61!. LT3!. 1/381!. 1/171!. 1/171!. 1/433!. 1/353!. 1/334!. 1/633!. 1/615!. 1/517!. 211!. LT!. 1/383!. 1/172!. 1/171!. 1/433!. 1/272!. 1/233!. 1/637!. 1/672!. 1/569!. 211!. LT2!. 1/383!. 1/171!. 1/171!. 1/433!. 1/269!. 1/245!. 1/637!. 1/652!. 1/548!. 211!. LT3!. 1/383!. 1/171!. 1/171!. 1/433!. 1/34:!. 1/32:!. 1/637!. 1/651!. 1/576!. 57. 1/661!.

(58) ! ϩᜪ҅ዴ౗. ˃ˁˈ ˃ˁˇ ˃ˁˆ ˃ˁ˅ ˃ˁ˄ ˃. ˥̆̀ˀ˞˦ ˪̅̆̀ˀ˞˦˄ ˪̅̆̀ˀ˞˦˅ ˕˚ˀ˥̆̀ˀ˞˦. ! &. ˕˚ˀ˪̅̆̀ˀ˞˦˄ ̄˷˶. ˾́́˶. ̆̉˶. ˕˚ˀ˪̅̆̀ˀ˞˦˅. ! ϩᜪᏔ კ 4-54 ௲‫ػ‬ෳᡍၗ਑໣ϩᜪ҅ዴ౗Кၨკ(ჴᡍ 9, Ni=10,B=20) ϩᜪ҅ዴ౗. ˃ˁˉ. ˥̆̀ˀ˞˦. ˃ˁˇ. ˪̅̆̀ˀ˞˦˄ ˪̅̆̀ˀ˞˦˅. ˃ˁ˅. ˕˚ˀ˥̆̀ˀ˞˦. ! &. ˕˚ˀ˪̅̆̀ˀ˞˦˄. ˃ ̄˷˶. ˾́́˶. ̆̉˶. ˕˚ˀ˪̅̆̀ˀ˞˦˅. ! ϩᜪᏔ კ 4-55 ௲‫ػ‬ෳᡍၗ਑໣ϩᜪ҅ዴ౗Кၨკ(ჴᡍ 9, Ni=10,B=50) ϩᜪ҅ዴ౗. ˃ˁˉ. ˥̆̀ˀ˞˦ ˪̅̆̀ˀ˞˦˄. ˃ˁˇ. ˪̅̆̀ˀ˞˦˅. ˃ˁ˅. ˕˚ˀ˥̆̀ˀ˞˦. ! &. ˕˚ˀ˪̅̆̀ˀ˞˦˄. ˃ ̄˷˶. ˾́́˶. ̆̉˶. ˕˚ˀ˪̅̆̀ˀ˞˦˅. ! ϩᜪᏔ კ 4-56 ௲‫ػ‬ෳᡍၗ਑໣ϩᜪ҅ዴ౗Кၨკ(ჴᡍ 9, Ni=10,B=100) ! ! !. 58.

(59) ! ߄ 4-10 ௲‫ػ‬ෳᡍၗ਑໣ II (Ni=20)ϩᜪ҅ዴ౗ Dmbttjgjfs! C!. Bmhpsjuin!. red! CH!. XSTN!. lood)l>2*! CH.!. CH!. XSTN!. XSTN!. twd! CH.!. CH!. XSTN!. XSTN!. 1/562!. CH.! XSTN!. 2!. TJOHMF!. 1/465!. 31!. LT!. 1/523!. 1/282!. 1/195!. 1/44:!. 1/351!. 1/329!. 1/715!. 1/524!. 1/433!. 31!. LT2!. 1/523!. 1/271!. 1/292!. 1/44:!. 1/379!. 1/354!. 1/715!. 1/5:8!. 1/512!. 31!. LT3!. 1/523!. 1/421!. 1/222!. 1/44:!. 1/43:!. 1/423!. 1/715!. 1/594!. 1/4:6!. 61!. LT!. 1/561!. 1/439!. 1/343!. 1/461!. 1/388!. 1/365!. 1/715!. 1/533!. 1/437!. 61!. LT2!. 1/561!. 1/34:!. 1/33:!. 1/461!. 1/39:!. 1/371!. 1/715!. 1/57:!. 1/489!. 61!. LT3!. 1/561!. 1/424!. 1/2:2!. 1/461!. 1/456!. 1/432!. 1/715!. 1/613!. 1/527!. 211!. LT!. 1/563!. 1/421!. 1/379!. 1/463!. 1/3:9!. 1/388!. 1/717!. 1/592!. 1/4:5!. 211!. LT2!. 1/563!. 1/424!. 1/388!. 1/463!. 1/427!. 1/3:9!. 1/717!. 1/625!. 1/617!. 211!. LT3!. 1/563!. 1/424!. 1/334!. 1/461!. 1/473!. 1/449!. 1/717!. 1/648!. 1/633!. ! ! !. 59. 1/53:!.

(60) ! ϩᜪ҅ዴ౗. ˃ˁˉ. ˥̆̀ˀ˞˦ ˪̅̆̀ˀ˞˦˄. ˃ˁˇ. ˪̅̆̀ˀ˞˦˅. ˃ˁ˅. ˕˚ˀ˥̆̀ˀ˞˦. ! &. ˕˚ˀ˪̅̆̀ˀ˞˦˄. ˃ ̄˷˶. ˾́́˶. ̆̉˶. ˕˚ˀ˪̅̆̀ˀ˞˦˅. ! ϩᜪᏔ კ 4-57 ௲‫ػ‬ෳᡍၗ਑໣ϩᜪ҅ዴ౗Кၨკ(ჴᡍ 10, Ni=20,B=20) ϩᜪ҅ዴ౗. ˃ˁˉ. ˥̆̀ˀ˞˦ ˪̅̆̀ˀ˞˦˄. ˃ˁˇ. ˪̅̆̀ˀ˞˦˅. ˃ˁ˅. ˕˚ˀ˥̆̀ˀ˞˦. ! &. ˕˚ˀ˪̅̆̀ˀ˞˦˄. ˃ ̄˷˶. ˾́́˶. ̆̉˶. ˕˚ˀ˪̅̆̀ˀ˞˦˅. ! ϩᜪᏔ კ 4-58 ௲‫ػ‬ෳᡍၗ਑໣ϩᜪ҅ዴ౗Кၨკ(ჴᡍ 10, Ni=20,B=50) ϩᜪ҅ዴ౗. ˃ˁˉ. ˥̆̀ˀ˞˦ ˪̅̆̀ˀ˞˦˄. ˃ˁˇ. ˪̅̆̀ˀ˞˦˅. ˃ˁ˅. ˕˚ˀ˥̆̀ˀ˞˦. ! &. ˕˚ˀ˪̅̆̀ˀ˞˦˄. ˃ ̄˷˶. ˾́́˶. ̆̉˶. ˕˚ˀ˪̅̆̀ˀ˞˦˅. ! ϩᜪᏔ კ 4-59 ௲‫ػ‬ෳᡍၗ਑໣ϩᜪ҅ዴ౗Кၨკ(ჴᡍ 10, Ni=20,B=100). 60.

(61) ӧҁࣴ‫ز‬྽ύǴ‫܌‬ගр‫ޑ‬ཥӭख़ϩᜪᏔǴbagging ᆶ୏ᄊᒧ‫ڗ‬ηᆢࡋ‫ ޑ‬WRSM ϐ่ӝᄽᆉ‫ݤ‬Ҕٰϩᜪଯᆢࡋၗ਑Ǵ٠٬ҔΟঁόӕਏ݀‫୷ޑ‬ᘵϩᜪᏔǴჴᡍ΢ аόӕ‫ޑ‬ϩᜪᏔঁኧϷ૽ግኬҁኧٰᡍ᛾‫܌‬ගрБ‫ޑݤ‬ਏૈǶҁࣴ‫ز‬ว౜ӵΠ‫܌‬ Ң:. 1.. ჴᡍ่݀ᡉҢа BG-WRSM-KS2 ᄽᆉ‫ݤ‬ӧ Washington DC Mall ᆶ Indian Pine Site ၗ਑໣΢Ҕ‫ܭ‬ϩᜪᏔ knnc(k=1)‫ ک‬BG-WRSM-KS1 ᄽᆉ‫ݤ‬Ҕ‫ܭ‬ϩᜪᏔ qdc ӧ Indian Pine Site ၗ਑໣ࢂᛙ‫ۓ‬Ъε೽ҽ௃‫ڀ׎‬Ԗၨ٫‫ޑ‬ਏૈ߄౜Ƕ. 2.. ӧ Washington DC Mall ၗ਑໣ύǴϩᜪᏔ qdcǵsvc ߄౜ᆶ WRSM ϩᜪ҅ዴ ౗εठ࣬ӕǹӧ Indian Pine Site ၗ਑໣ύ߾ࢂϩᜪᏔ svc ߄౜ᆶ WRSM ϩᜪ ҅ዴ౗εठ࣬ӕǶ. 3.. ӧ Indian Pine Site ၗ਑໣ύǴว౜྽૽ግኬҁኧၨϿਔǴϩᜪᏔঁኧຫӭ߾ ჹϩᜪ߄౜཮Ԗၨᡉ๱ቹៜǴԶ૽ግኬҁኧၨӭਔǴ߾ҔၨϿ‫ޑ‬ϩᜪᏔঁኧ ջёၲ‫ډ‬ന٫ॶǹԶӧ Washington DC Mall ၗ਑໣ύǴ૽ግኬҁኧ‫ޑ‬ӭჲᆶ ϩᜪᏔঁኧ‫߄ޑ‬౜่݀คϼεৡ౦Ƕ. 4.. ӧჴᡍၸำύǴӢࣁ‫܌‬٬Ҕ‫ޑ‬ϩᜪᏔঁኧКচҁ٬Ҕ‫ޑ‬Б‫ ݤ‬bagging ᆶ WRSM ٰ‫ޑ‬ӭǴӢԜӧ૽ግၸำ‫ޑ‬ਔ໔΢཮Кচҁ‫ٿޑ‬ᅿБ‫ݤ‬೿ाٰ‫ޑ‬ΦǴ ೭ࢂҁࣴ‫ز‬Ѹ໪‫ׯ‬຾‫ޑ‬લᗺϐ΃Ǵՠϩᜪ҅ዴ౗΢ዴԖܴᡉ‫ׯޑ‬๓Ƕ. 5.. ௲‫ػ‬ෳᡍၗ਑໣ύค‫ݤ‬Ԗၨӳ‫ׯޑ‬๓‫ޜ‬໔ࢂҗ‫ঁ؂ܭ‬ᚒҞȐᆢࡋȑ֡ࢂҗ஑ ৎ‫܌‬೛ीၸ‫ޑ‬Ǵ‫܌‬аΨଞჹᚒҞ‫཮܌‬೷ԋ‫ޑ‬ᒱᇤᜪࠠуа᏾ӝǴ‫ঁ؂ࢂܭ‬ᚒ Ҟ໔‫܄ޑ‬፦ࢂឦ‫ܭ‬όёϩപ‫ޑ܄‬ǶӢԜӵ݀٬Ҕ WRSM Б‫ڗܜٰݤ‬ᆢࡋ཮ Ԗᝄख़ཞѨૻ৲ໆ‫่݀ޑ‬วғǴω཮೷ԋၗ਑ϩᜪ่݀දၹό౛གྷ‫ݩރޑ‬Ƕ. 61.

(62) ಃϖക!่ፕ‫ک‬҂ٰว৖! ӧҁࣴ‫ز‬ύǴନΑှ، WRSM ѝൂჹᆢࡋբ‫ڗܜ‬ୢᚒϐѦǴᗋуΕჹኬҁբ ‫ޑڗܜ‬Б‫ݤ‬Ǵࡌᄬ΋঺ૈ୼ӕਔှ،ኬҁᒧ‫ڗ‬ᆶԾ୏ᒧ‫ڗ‬ᆢࡋ‫ޑ‬ӭख़ϩᜪᏔ‫س‬ ಍Ǵ٠ჹ٬ҔচБ‫֡ޑݤ‬΋ϩଛǵ૽ግኬҁϩᜪ҅ዴ౗Ϸጕ‫୔܄‬ձϩ‫ޑ݋‬ϩᚆໆ ‫ࡌ܌‬ᄬ‫ޑ‬ᆢࡋϩଛ଺КၨǴࣁКၨόӕϩᜪᏔ‫ޑ‬ϩᜪਏૈǴճҔ 3 ᅿόӕ‫ޑ‬ϩᜪ ᏔǴ٠ჹόӕ‫૽ޑ‬ግኬҁኧϷϩᜪᏔঁኧ଺૸ፕǴᔠຎ‫ځ‬ύ‫ޑ‬ৡ౦‫܄‬Ϸ‫ځ‬ਏૈࣁ ՖǶ൩ჴᡍ่݀ᡉҢрǴҁࣴ‫܌ز‬ගϐБ‫ݤ‬Ǵӧ Washington DC Mall ၗ਑໣ύǴ qdc Ϸ svc ϩᜪᏔᆶচБ‫ݤ‬ਏૈ࣬՟ǴԶӧ knnc ߾ૈຬຫচБ‫߄ޑݤ‬౜Ƕӧ Indian Pine Site ၗ਑໣ύǴsvc ϩᜪᏔᆶচБ‫ݤ‬ਏૈ࣬՟ǴԶӧ qdc Ϸ knnc ߾ૈຬຫচ Б‫߄ޑݤ‬౜Ƕӧόӕ‫૽ޑ‬ግኬҁኧϷϩᜪᏔঁኧ‫ޑ‬ቹៜǴёаว౜‫ډ‬ӧ Indian Pine Site ၗ਑໣ύǴ྽૽ግኬҁኧၨϿਔǴϩᜪᏔঁኧຫӭ߾ჹϩᜪ߄౜཮Ԗၨᡉ๱ ቹៜǴԶ૽ግኬҁኧၨӭਔǴ߾ҔၨϿ‫ޑ‬ϩᜪᏔঁኧջёၲ‫ډ‬ന٫ॶǹԶӧ Washington DC Mall ၗ਑໣ύǴ૽ግኬҁኧ‫ޑ‬ӭჲᆶϩᜪᏔঁኧ‫߄ޑ‬౜่݀คϼ εৡ౦Ƕ‫ࢂܭ‬ёаள‫ޕ‬ሡຎኬҁ‫ޑ‬௃‫׎‬Ǵჹ૽ግኬҁኧ‫ޑ‬ελٰፓ᏾ϩᜪᏔঁ ኧǶԶӧӚϩᜪᏔύ knnc ёаள‫ډ‬ၨᡉ๱‫ׯޑ‬๓ਏૈǶ வ Washington DC Mall आѦጕ‫ܡޜ‬კύǴёаၨܴᡉϩрᜪձ‫୔ޑ‬༧ǴԶӧ Indian Pine Site आѦጕ‫ܡޜ‬კύǴၨόܰϩᒣрᜪձ‫ޑ‬ϩഁǴҗࣴ‫ز‬ύว౜Ǵҁ ࣴ‫܌ز‬ගрϐБ‫ݤ‬ёаஒၨܰషᚇ‫ޑ‬ၗ਑ᜪࠠǴբрၨԖਏ‫ׯޑ‬๓ਏૈǴӢԜё аᇥၗ਑ᜪࠠऩࢂၨܰషᚇᜤᒣ‫ޑ‬Ǵ߾ҁࣴ‫܌ز‬ගϐБ‫ࢂݤ‬ၨᛙ଼Ъ‫ڀ‬Ԗ‫ׯ‬๓ਏ ૈ‫ޑ‬Ƕ ӧ҂ٰࣴ‫ز‬ᆶว৖Бय़Ǵࡌ᝼ёаჹϩᜪᏔঁኧ‫ޑ‬ελ‫ک‬ኬҁኧελǴϷჹ όӕ‫ޑ‬ϩᜪᏔ่ӝ౛ፕ଺‫׳‬຾΋‫૸ޑ؁‬ፕǴќѦᗋёауΕόӕ‫ޑ‬੝ቻу៾Б ԄǴаϷ٬Ҕόӕ‫ޑ‬ኬҁ‫ڗܜ‬БԄ‫ޣ܈‬٬Ҕ‫׳‬ӭኬҁኧٰ຾Չ૽ግϩᜪᏔǴќѦ. 62.

(63) Եቾჹኬҁࡌҥќ΋ঁჹኬҁ‫ޑڗܜ‬ϩଛǴ٠Ҕ‫ځ‬д‫ޑ‬у៾Б‫׳ٰݤ‬ཥ‫ޑڗܜ‬ϩ ଛǴаԜБ‫ڗܜٰݤ‬ኬҁ཮Кၨ࠼ᢀЪᛙ଼Ǵ‫܈‬೚ૈගଯϩᜪᏔ‫ޑ‬ਏૈǶ. 63.

(64)

參考文獻

立即下載 ( PDF - 63 頁 - 0.94 MB )

相關文件

The Joy of Sampling

For ex- ample, if every element in the image has the same colour, we expect the colour constancy sampler to pro- duce a very wide spread of samples for the surface

of the Atom

You are given the wavelength and total energy of a light pulse and asked to find the number of photons it

The Importance of the Rule of Law to Hong Kong SAR

(1) principle of legality - everything must be done according to law (2) separation of powers - disputes as to legality of law (made by legislature) and government acts (by

7.1. The Ring of Integers

好了既然 Z[x] 中的 ideal 不一定是 principle ideal 那麼我們就不能學 Proposition 7.2.11 的方法得到 Z[x] 中的 irreducible element 就是 prime element 了..

The hidden dimensions of the Universe

volume suppressed mass: (TeV) 2 /M P ∼ 10 −4 eV → mm range can be experimentally tested for any number of extra dimensions - Light U(1) gauge bosons: no derivative couplings. =&gt;

2 Formation of the FISDW

For pedagogical purposes, let us start consideration from a simple one-dimensional (1D) system, where electrons are confined to a chain parallel to the x axis. As it is well known

of the QCDVacuum

incapable to extract any quantities from QCD, nor to tackle the most interesting physics, namely, the spontaneously chiral symmetry breaking and the color confinement.. 

The Dark Ages of the Universe

• Formation of massive primordial stars as origin of objects in the early universe. • Supernova explosions might be visible to the most