2, 84 o. 2, 2010 2010 6 15 R EM O TE SE S I G FO R LAD & RESOU RC ES Jun., 2010 AdaBoost 1, 2, 4, 2, 2, 3 (1., 475000; 2., 100101; 3., 730070; 4. 75711, 510515) : 2007 CBERS, AdaBoost,,, AdaBoost : AdaBoost; ; ; CBERS : TP 701 : A : 1001-070X (2010) 02-0086 - 05 0,,,,,, C. P. Lo [ 1 ] 70 mm,, Landsat, 1 2. 5,, [ 2 ] Landsat TM, [ 3 ] (DB I), TM,,, [ 4 ] Landsat TM,, DV I DW I,, J. D. Paola [ 5 ] J. Gao [ 6 ] SPOT,,,,,,,, [ 7 ],, [ 8 ] [ 9 ] [ 10 ],, AdaBoost, AdaBoost : 2009-12 - 25; : 2010-02 - 22
2, : AdaBoost 87 [ 11 ] [ 12 ],,, MRF [ 13 ] [ 14 ] ASTER, AdaBoost -,,, [ 15 ], Quick B ird, DV I, Adaboost, Quick B ird,, AdaBoost, 1 AdaBoost 1. 1 Boosting,,,,, Boosting,, Boosting AdaBoost Boos2 ting, Boosting, Y. Freund 1995 [ 18 ],, Boosting [ 19 ] AdaBoost,,,, ;,,, AdaBoost,, AdaBoost,,, [ 14 ] 1. 2,, AdaBoost [ 20 ] :, S = { ( x 1, y 1 ),, ( x, y ) } (1), y i { + 1, - 1},, + 1, - 1 AdaBoost, ( x i, y i ) t d t i ( ),,,, t ( h t, d ( t) ) = i I[ y i h t ( x i ) ] (2) i = 1 d( t) h t, t, G ( t ) = exp{ - y i [ t j h j ( x j ) ] } (3) i = 1 j = 1, H T ( x) = T t h t ( x) (4), H ( x) = sing [ H T ( x) ] (5), [ 19, 20 ] : S = { ( x 1, y 1 ), ( x 2, y 2 ),, ( x, y ) }, y i { + 1, - 1} (6) d ( t) i = 1, i = 1, 2,, (7),, T, T,
88 2010 2 h t xϖ { + 1, - 1} (8) h t t, : t = i I[ y i h t ( x i ) ] (9) i = 1 d( t) lg 1 - t d ( t + 1) i = d ( t) i t d t + 1 i = d exp [ - t y i h t ( x i ) ] (10) ( t + 1) i / t + 1) i = 1 d( i (11) t =0 t 1,, T = t - 1, 2, t t / t (12) = H ( x) = sign [ T t h t ( x) ] (13) 3 2 AdaBoost,, 3 2. 1, 5, 1 [ 23 ] 1 CBERS F ig. 1 The spectrum respon se curves of urban area by CBERS image 2. 2, p ( i, j), (4) ( Contrast), ( GLCM ),, L - 1 L - 1L (1) ( Entropy) f 4 = n = 0 n2 { - 1 p^ ( i, j) } i - j = n (17) i = 0 j = 0, (5) (Homogeneity) L - 1L f 1 = - - 1 p^ ( i, j) lbp^ ( i, j) (14) i = 0 j = 0, (2) (Angular Second Moment or Energy) L - 1L f, 5 = - 1 p^ ( i, j) (18) i = 0 j = 0 1 + ( i - j) 2 L - 1L - 1 f 2 = p^ 2 ( i, j) (15) i = 0 j = 0 (3) (Variance) f 3 L - 1L - 1 = [ p^ ( i, j) ( i - ) 2 ] (16) i = 0 j = 0 1 5 5, 1 Tab. 1 The texture characters of urban land use cover and non - urban land use cover of CBERS 000. 096 0 06. 058 9 5. 218 8 0. 351 2 002. 201 9 06. 216 4 05. 504 6 0. 257 0-0. 610 4 00. 943 2 0. 210 1 0. 206 1-0. 566 4 00. 923 2 00. 488 6 0. 192 0 00. 0 12. 000 0 8. 059 0 2. 377 4 012. 001 2 79. 710 4 19. 920 1 7. 998 0 00. 0 02. 888 9 0. 281 9 0. 160 3 00 00 04. 500 0 00. 537 4 0. 318 9 00. 0 01. 000 0 0. 475 8 0. 222 1 00 00 01. 000 0 00. 242 9 0. 106 0
2, : AdaBoost 89 2. 3 CBERS - 02 CCD, CBERS, DB I,, ;, ; DV I DW I ; ; [ 23 ] 2, 2 F ig. 2 Flow chart of dec ision tree cla ssif ier a lgor ithm 3 CBERS, ; AdaBoost 3 ; 3. 1, 30 08 20 31 23 22, 121 27 18 121 48 43, CBERS - 02 CCD, 2007 5 26, Path366 /Row65, 4 3 2 3. 2 AdaBoost CBERS ( 1) ;,, 1, 2 ( 19) 3, ( 3) H ( x) = sign [ T t h t ( x) ] = 1 0 (19) 3 F ig. 3 Image of study area and respective cla ssif ica tion result of each cla ssif ier
90 2010 3. 3 2, Adaboost Adaboost 11. 91%, 27. 08% 14. 06%, 17. 68%, Adaboost 2 Tab. 2 The exper im en t result Adaboost 68. 212 5 53. 038 0 66. 061 8 80. 123 9 4 AdaBoost,, AdaBoost,, : [ 1 ] Lo C P. Land U se Mapp ing of Hong Kong From Landsat Images: An Evaluation[ J ]. International Journal of Remote Sensing, 1981, 2 (3) : 231-252. [ 2 ],. TM [ J ]., 2000, 4 (2) : 146-150. [ 3 ],,. TM [ J ],. 2003, 7 (1) : 37-40. [ 4 ],,,. TM [ J ]., 2006, 21 (1) : 31-36. [ 5 ] Paola J D, Schowengerdt R A. A Detailed Comparison of Back2 p ropagation eural etwork and Maximum - likelihood Classifiers for U rban Land U se Classification[ J ]. IEEE Transactions on Geo2 sciences and Remote Sensing, 1995, 30 (4) : 981-996. [ 6 ] Gao J, Skillcorn D. Capability of SPOT XS Data in Producing De2 tailed Land CoverMap s at the U rban - rural Periphery[ J ]. Inter2 national Journal of Remote Sensing, 1998, 19 (15) : 1366-5901. [ 7 ] Dunrpw T. Experiments with Classifier Combing Rules[ C ] / /Proc of the 1st International Workshop on Multip le Classifier System s, Cagliari, Italy, 2000: 16-29. [ 8 ] Gutta S, W echsler H. Face Recognition U sing Hybrid Classifier System s[ C ] / / Proc of IEEE International Conference on eural etworks,w ashington DC, 1996. [ 9 ] H ansen L K, L iisber G L, Salamon P. Ensem ble M ethods for Handwritten D igit Recognition [ C ] / / Proc of IEEE Workshop on eural etworks for Signal Processing, Copenhagen, 1992. [ 10 ] ZHOU Zhi - hua, J IAG Yuan, YAG Yu - bin, et al. Lung Cancer Cell Identification Based on A rtificial eural etwork En2 sembles[ J ]. A rtificial Intelligence in Medicine, 2002. 24 ( 1) : 25-36. [ 11 ] ishii R, Eguchi S. Supervised Image Classification by Contextual AdaBoost Based on Posteriors in eighborhoods[ J ]. IEEE Trans2 actions on Geoscience and Remote Sensing, 2005, 11 ( 43) : 2547-2554. [ 12 ] ishii R, Eguchi S. Supervised Image Classification Based on Ada2 Boost with Contextual W eak Classifiers[ J ]. Geoscience and Re2 mote Sensing Symposium, 2004, 9 (2) : 1467-1470. [ 13 ] Vapnik V. The ature of Statistical Learning Theory[M ]. ew York: Sp ringer, 2000. [ 14 ],,,. AdaBoost [ J ]., 2007, 24 ( 10) : 181-184. [ 15 ],,. AdaBoost [ J ]., 2008, 44 (20) : 23-25. [ 16 ] Friedman J, Hastie T, Tibshirani R. Additive Logistic Regression: a Statistical V iew of Boosting (with D iscussion) [ J ]. Annals of Sta2 tistics, 2000, 28 (2) : 337-407. [ 17 ] Hastie T, Tibshirani R, Friedman J. The Elements of Statistical Learning: DataM ining, Inference, and Prediction[M ]. ew York: Sp ringer, 2001. [ 18 ] Freund Y, Schap ire R E. A Decision - theoretic Generalization of On - line Learning and an Application to Boosting[ J ]. Journal of Computer and System Science, 1997, 55 (1) : 119-139. [ 19 ] Robert E S. The Boosting Approach to Machine Learning: An O2 verview [ C ] / /Proceedings of MSR I Workshop on onlinear Esti2 mation and Classification, 2002. [ 20 ]. [ D ]., 2006. [ 21 ],. [ J ]., 2002 (1) : 27-32. [ 22 ],,. [ J ]., 2006, 21 (6) : 159-162. [ 23 ],,,. CBERS - 02 CCD [ J ]., 2008 (7) : 504-509. [ 24 ] ishii R, Eguchi S. Robust Supervised Image Classifiers by Spatial AdaBoost based on RobustLoss Functions[ C ] / / Proc of SPIE, Im2 age and Signal Processing for Remote Sensing X I, 2005. [ 25 ] Zha Y J, Gao S. U se of ormalized D ifference Built - up Index in Automatically Mapping U rban A reas from TM Imagery[ J ]. In2 ternational Journal of Remote Sensing, 2003, 24 (3) : 583-594. [ 26 ] Gao J, Skillcorn D. Capability of SPOT XS Data in Producing De2 tailed Land CoverMaps at the U rban rural Periphery[ J ]. Interna2 tional Journal of Remote Sensing, 1998, 19 (6) : 2877-2891. ( 96 )
96 2010 Rem ote Sen sing Ba sed Spa tio - tem pora l Evolution of Land Use Pa ttern in Huangpu R iver Coa st FEG Yong - jiu, HA Zhen ( College of M arine Sciences, Shanghai O cean U niversity, Shanghai 201306, China) Abstract: Land use and cover changes are remarkably characterized by the temporal evolution and spatial differences. Comp rehensive studies of these characteristics are very important for understanding the structure of land use and the trend of urbanization. W ith the support of the remote sensing and geographical information system ( GIS) techniques, the authors incised a width of 5km along the Huangpu R iver coast as the study area from the Landsat TM /ETM + images of Shanghai in 1992 and 2008. Moreover, for a comparative analysis, the study area was segmented into four sub2areas w ith the inner and exterior ring roads of Shanghai C ity. classified by using the m inimum distance classifier in Envi 4. 2 to map the land use. The two images were Several indexes, such as land use dynam ic degree, intensity and relative change rate, were calculated to describe the land use changes. W ith the spatial analysis functions for the graphic in a GIS environment, new ly increased built2up area were mapped to qualitatively analyze the spatio2temporal evolution of land use in Huangpu R iver coast. that the built2up area was greatly increased, This study has demonstrated the gardened area was slightly raised, and the cultivated land was drastically decreased in Huangpu R iver coast form 1992 to 2008. A t the same time, the land use dynam ic degree and the intensity of the eastern area were larger than the indexes of the average values of the whole Huangpu R iver coast. Key words: Land use change; Remote sensing; Spatio - temporal evolution; Huangpu R iver coast : (1981 - ),,,, GIS ( : ) ( 90 ) An Adaboost Ba sed M ethod for D ynam ic Extraction of Urban Land Use Cover w ith Re m ote Sen sing Images L I Rui 1, 2, 4, WAG Juan - le 2, RE Zheng - chao 2, 3 ( 1. College of Environm ent and Planning, Henan U niversity, Kaifeng 475000, China; 2. S tate Key Laboratory of Resources and Environm ental Inform ation System, Institute of Geographical Sciences and atural Resources Research, CAS, B eijing 100101, China; 3. College of grass industry, Gansu Agricultural University, Lanzhou 730070, China; 4. The 75711 Troop, PLA, Guangzhou 510515, China) Abstract: The p roblem how to combine the low p recision urban land use cover classifiers to get higher p recision is dealt w ith in this paper. U sing 2007 Shanghai CB ERS ( China - B razil Earth Resources Satellite) images, the authors adop ted the AdaBoost combination classifier, which can com bine spectral feature information, texture structure information and decision tree classier to imp rove the classification p recision. The experiment results show that a notable imp rovement of the classification p recision of urban land use cover can be achieved by using AdaBoost algorithm. Key words: AdaBoost; U rban land use cover; Remote sensing image; CBERS : (1979 - ),,, ( : )