臺灣博碩士論文加值系統

理論上，不同種類的植物擁有不同的光譜特徵，前人研究報告主要以土地利用型之分類為重點，較少的研究是討論樹種間光譜特徵變化，而用遙測影像判釋植生種類，因有許多的限制因子影響了樹種判釋準確度，故較分辨土地利用型困難。本研究採用捷鳥(QuickBird)衛星影像為研究材料，因其為高解析力影像，可讓像元較為單純進而使樹木冠層影像單純化，增加樹木冠層像元資料之可用價值。
本研究的目的在於以衛星影像來分辨樹種，用QB全色態及多光譜影像採用了灰階共現矩陣(GLCM)組織特徵萃取、植生指標、Gamma、Lee-sigma及Mean組織濾波器等方法，將原本5波段之QB影像延展為18波段的影像，以資料延展的概念提升樹種分辦能力，而因延展後18波段影像做樹種分類時，會有Hughes現象的發生，因此利用主軸轉換法來避免之。
QB多光譜影像分類結果全區分類準確度(OA)值60.61%， 值0.5915，區外分類準確度9.18%；GLCM組織特徵萃取以MEAN+HOM +CON+DIS+ENT的分類結果會最佳，OA值29.61%， 值0.2750，區外分類準確度3.83%；植生指標以ARVI+EVI+NDVI+SAVI+VARI的分類結果最佳，OA值50.86%及 值0.4922，區外分類準確度7.72%；Gamma濾波器以33*33視窗大小處理之效果最佳，OA值98.66%及 值0.9859，區外分類準確度8.98%；Lee-sigma濾波器以39*39視窗大小處理之效果最佳，OA值98.06%及 值0.9797，區外分類準確度6.93%；Mean濾波器以39*39視窗大小處理之效果最佳，OA值98.95%及 值0.9890，區外分類準確度10.38%。
訓練樣本大小及濾波器處理視窗大小關係的研究中，以Gamma濾波器處理之結果，經平均計算後得到最適視窗大小約為樹種訓練樣本長的1.78±0.80倍，寬的1.81±0.84倍；以Lee-sigma濾波器處理之結果，約為樹種訓練樣本長的2.03±1.10倍，寬的2.07±1.19倍；以Mean濾波器處理之結果，約為樹種訓練樣本長的1.62±0.82倍，寬的1.68±1.00倍，因標準偏差過大，顯示訓練樣本與最適視窗大小並無明顯的關係存在。而以三種濾波器處理得到之各樹種最適視窗做比較， Gamma vs. Lee-sigma之相關係數為0.5611，Lee-sigma vs. Mean之相關係數為0.5277，Mean vs. Gamma之相關係數為0.6511，相關性為中度相關，所以並非僅做一種濾波器，得到訓練樣本之最適視窗，即可套用到其他的濾波器處理影像的結果。
主軸轉換後影像以相鄰特性根閾值法決定主軸數，並以不同分類法做比較，結果發現區內分類準確度評估以最大概似法為最好，最佳之結果為結合Mean組織特徵，完全分類正確沒有任何誤分的情形發生，而在區外分類準確度評估以Mahalanobis距離分類法的效能最佳，最佳之結果為結合Lee-sigma組織特徵，OA值11.85%。

Logically, plants have their own particular spectral characteristics which could be applied for tree classification with remote sensing techniques. Unfortunately, most of former researches focus on study landuse classification and have demonstrated that it is not easy to clearly recognize forest vegetation. Spatial information has been proven to be workable in the classification of land covers recently. Tree species identification is more critical and difficult than land covers. A lot of factors affect accurate of tree species identification, but we try to solve the problem. This paper attend to emphasize the potentially valuable signals of remote sensing to detect tree species using a higher spatially resolution of QuickBird images.
The purpose of this paper is identifying tree species using satellite image. We adopt Gray Level Co-occurrence Matrix, Vegetation Index, Gamma texture filter, Lee-sigma texture filter and Mean texture filter to extract new information promote classification accuracy. Than, original five bands image become eighteen bands image to increase more information to tree classification. However, when we apply new image to perform our research and occur Hughes phenomenon, so employ Principle Component Analysis to avoid Hughes phenomenon.
The classified accuracy of QB spectral image performs overall accuracy 60.61%, Kappa hat 0.5915 and assessing accuracy 9.18%. The best combination is MEAN+HOM+CON+DIS+ENT using GLCM method, performs classified result overall accuracy 29.61%, Kappa hat 0.2750 and assessing accuracy 3.83%. The best combination is ARVI+EVI+NDVI+SAVI+VARI using Vegetation Index method, performs classified result overall accuracy 50.86%, Kappa hat 0.4922 and assessing accuracy 7.72%.Using Gamma filter and window size 33*33 processes QB image is the best result, performs classified result overall accuracy 98.66%, Kappa hat 0.9859 and assessing accuracy 8.98%. Using Lee-sigma filter and window size 39*39 processes QB image is the best result, performs classified result overall accuracy 98.06%, Kappa hat 0.9797 and assessing accuracy 6.93%. Using Mean filter and window size 39*39 processes QB image is the best result, performs classified result overall accuracy 98.95%, Kappa hat 0.9890 and assessing accuracy 10.38%.
We also study relationship between training sample size and filter process window size. Processing QB-mss image using Gamma filter, each tree species optimal widow size approximate 1.78±0.80 times of row pixel of training sample and 1.81±0.84 times of column pixel of training sample. Processing QB-mss image using Lee-sigma filter, each tree species optimal widow size approximate 2.03±1.10 times of row pixel of training sample and 2.07±1.19 times of column pixel of training sample. Processing QB-mss image using Mean filter, each tree species optimal widow size approximate 1.62±0.82 times of row pixel of training sample and 1.68±1.00 times of column pixel of training sample. It indicates that training sample and optimal widow size have not significant relationship. Than we compare each tree species optimal widow size among three filter processing result, obtain correlation coefficient of Gamma vs. Lee-sigma, Lee-sigma vs. Mean and Mean vs. Gamma, 0.5611, 0.5277 and 0.6511 respectively. It indicates that using different filter processing image, optimal window size of each training samples is different.
After using principal component analysis, we adopt threshold of slope of adjacent eigenvalues method to determine band number, and compare the classified accuracy of using different methods of image classification. Results show that classified accuracy of training samples using maximum likelihood classifier is the best, and combine Mean texture characteristic the Kappa hat is 1.0. Using assessing samples to evaluate classified accuracy and we find combine Lee-sigma texture characteristic to evaluate classified accuracy using Mahalanobis distance classifier is the best, and overall accuracy is 11.85%.

圖次
表次
中文摘要
英文摘要
第一章 緒論 1
1-1 研究動機與目的 1
1-2論文架構 3
第二章 植生指標 4
2-1 以斜率為基準之植生指標 5
2-1-1 抵抗大氣植生指標 5
2-1-2 規整差植生指標 5
2-1-3 可見光段抵抗大氣植生指標 6
2-2 以距離為基準之植生指標 6
2-2-1 增揚植生指標 7
2-2-2 土壤修正植生指標 7
第三章 組織濾波器 9
3-1 灰階共現矩陣 9
3-2 Gamma組織濾波器 11
3-3 Lee-sigma組織濾波器 11
第四章 影像分類法 13
4-1 光譜分離度 13
4-2 最短距離分類法 15
4-3 最大概似分類法 15
4-4 Mahalanobis距離分類法 16
4-5光譜角分類法 17
4-6 分類精度評估方法 17
第五章 光譜轉換 20
5-1 主成份分析 20
5-2 應用PCA法以避免Hughes現象 21
5-3 主軸數決定方法 22
5-3-1 相鄰特性根斜率閾值法 22
第六章 研究材料 24
6-1 QB衛星介紹 24
6-2 研究材料及試區概述 25
6-3 QB多譜態影像資料特性 26
6-3-1 植物 27
6-3-2 土壤及建築物 27
6-3-3 水體 27
第七章 分析及分類流程 29
7-1 建置立木GIS圖層 29
7-2 GLCM組織特徵萃取 30
7-3 多譜態影像分析 30
7-3-1 大氣糾正 30
7-3-2 引入植生指標及組織濾波器資訊 31
7-4 主軸轉換法 32
第八章 結果與討論 33
8-1 嘉大蘭潭校區立木GIS圖層 33
8-2 QB全色態影像GLCM組織特徵萃取 33
8-2-1 單一GLCM組織特徵 34
8-2-2 結合2個GLCM組織特徵 35
8-2-3 結合3個GLCM組織特徵 35
8-2-4 結合4個GLCM組織特徵 37
8-2-5 結合5個及6個GLCM組織特徵 37
8-2-6 GLCM組織特徵數目與分類精度之關係 38
8-3 QB多譜態影像分析 39
8-3-1 引入植生指標資訊 39
8-3-1-1 單一植生指標分類效能評估 39
8-3-1-2 結合2個植生指標分類效能評估 40
8-3-1-3 結合3個植生指標分類效能評估 41
8-3-1-4 結合4個及5個植生指標分類效能評估 41
8-3-1-5 植生指標數目與分類精度之關係 42
8-3-2 引入組織濾波器資訊 43
8-3-2-1 Gamma組織濾波器 43
8-3-2-1-1 最佳視窗處理大小 43
8-3-2-1-2 訓練樣本最適視窗及樣本大小之關係 52
8-3-2-2 Lee-sigma組織濾波器 57
8-3-2-2-1 最佳視窗處理大小 58
8-3-2-2-2 訓練樣本最適視窗及樣本大小之關係 61
8-3-2-3 Mean組織濾波器 67
8-3-2-3-1 最佳視窗處理大小 67
8-3-2-3-2 訓練樣本最適視窗及樣本大小之關係 70
8-3-3 小結 76
8-4 主軸轉換分析 77
8-4-1 加入Gamma濾波器之最佳處理結果 77
8-4-2 加入Lee-sigma濾波器之最佳處理結果 78
8-4-3 加入Mean濾波器之最佳處理結果 80
8-5 區外分類準確度評估 81
第九章 結論與建議 91
參考文獻 95

王慧君 (2004) 馬氏-田口系統之特性探討及其於信用評分之應用。國立交通大學工業工程管理系所博士論文。
王鑫 (1977) 遙測學。大中國圖書公司印行，P430。
沈明來 (1998) 實用多變數分析。九州圖書文物有限公司，P559。
林金樹 (1994) 光譜指標在環境變遷檢測上功效之研究。航空測量及遙感探測 26: 43-78。
林金樹 (1997) 動態學習法則類神經網路法在闊葉林區土地利用型分類上應用之研究。中華林學季刊 30(2): 151-164。
林金樹 (1999) 森林植生季節性光譜特性之研究。台灣林業科學14(3): 289-305。
林金樹 (2001) 影像參數加權統計法PN指標於土地利用型高光譜特徵選粹上之研究。航測及遙測學刊6(2): 1-20。
林金樹 (2002) 高光譜影像辨識土地利用型最適主軸數之決定。台灣林業科學 17(4): 471-481。
林金樹 (2003) 彩色航空照片數位影像在水稻田坵塊與面積調查應用上之研究。嘉義大學學報75: 181-198。
林金樹、管立豪 (2002) 利用航測與地理資訊系統技術調查阿里山針闊葉林自然保護區地形與林型分佈之研究。航測及遙測學刊7(1):35-52。
林務局 (1995) 第三次台灣森林資源及土地利用型調查。林務局出版，P222。
周朝富、鄭祈全 (1992) 利用SPOT衛星資訊探測林地覆蓋變遷之研究。林業試驗所研究報告 7(4): 305-317。
徐家盛 (2003) 多種尺度遙測影像於農業土地利用分類精確度探討。中華大學土木工程學系碩士論文。
浦瑞良、宮鵬 (2002) 高光譜遙測及其應用。五南圖書出版股份有限公司，P332。
郭麟霂 (2000) 寒帶沼地高光譜影像分類之研究。國立交通大學土木工程學系碩士論文。
葉志聖 (1995) AMOEBA影像分類技術之研究。國立交通大學土木工程學系碩士論文。
劉建慧 (1995) SPOT衛星影像之輻射改正。國立中央大學大氣物理研究所博士論文，P217。
魏曉萍 (2003) QuickBird衛星影像探討分類方法之研究。中華大學土木工程學系碩士論文。
QuickBird捷鳥衛星影像 (2003) http://www.quickbird.com.tw/
Arzandeh, S. and J. Wang (2002) Texture Evalution of Radarsat Imagery for Wetland Mapping. Canadian Journal of Remote Sensing 28:653-666.
Baraldi, A. and F. Parmiggiani (1995) An Investigation of the Textural Characteristics Associated with Gray Level Cooccurrence Matrix Statistical Parameters. IEEE Transactions on Geoscience and Remote Sensing 28:653-666.
Baret, F. and G. Guyot (1991) Potentials and Limits of Vegetation Indices for LAI and APAR Assessment. Remote sensing of Environment 35:161-173.
Chavez, J. P. (1988) An Improved Dark-object Subtraction Technique for Atmospheric Scattering Correction of Multispectral Data. Remote Sensing of Environment 24: 459-479.
Conese, C., M. J. Gilabert, F. Maselli and L. Bottai (1993) Topographic Normalization of TM Scenes Through the Use of an Atmospheric Correction Method and Digital Terrain Models. Photogram Engineering Remote Sensing 59(12): 1745-1753.
Conners, R. W. and C. A. Harlow (1980) A Theoretical Comparison of Texture Algorithms. IEEE Transactions on Pattern Analysis and Machine Intelligence PAMI-2:204-222.
Crippen, R. E. (1989) A Simple Spatial Filtering Routine for the Cosmetic Removal of Scanline Noise from Landsat TM P-tape Imagery. Photogrammetric Engineering and Remote Sensing 55(3):327-331.
Eastman, J. R. (1999) Guide to GIS and Image Processing: Volume2. Clark Labs. 170p.
Eliason, E. M. and A. S. McEwen (1990) Adaptive Box Filters for Removal of Random Noise from Digital Images. Photogrammetric Engineering and Remote Sensing 56(4):453-458.
Gitelson, A. A., Y. J. Kaufman, R. Starlc and D. Rundquist (2002) Novel Algorithms for Remote Estimation of Vegetation Fraction. Remote Sensing of Environment 80:76-87.
Gong, P., D. J. Marceau and P. J. Howarth (1992) A Comparison of Spatial Feature Extraction Algorithms for Land-use Classification with SPOT HRV Data. Remote Sensing of Environment 40:137-151.
Haralick, R. M., K. Shanmugam and I. Dinstein (1973) Textural Features for Image Classification. IEEE Transactions on System, Man and Cybernetics SMC-3:610-621.
Haralick, R. M. (1979) Statistical and Structural Approaches to Texture. Proceedings of the IEEE 67:786-804.
Herstein, I. N. and D. J. Winter (1988) A Primer on Linear Algebra. NY: Macmillan Publishing Company.
Hoffbeck, J. P. and D. A. Landgrebe (1996) Covariance Matrix Estimation and Classification with Limited Training Data. IEEE Transactions on Pattern Analysis and Machine Intelligence 18(7): 763-767.
Hudson, W. D. and C. W. Ramm (1987) Correct Formulation of the Kappa Coefficient of Agreement. Photogrammetric Engineering and Remote Sensing 3(4): 421-422.
Huete, A. R. (1988) A Soil-adjusted Vegetation Index (SAVI). Remote Sensing of Environment 25: 295-309.
Huete, A. R. and C. Justice (1999) MODIS Vegetation Index(MOD13) Algorithm Theoretical Basis Document. Greenbelt: NASA Goddard Space Flight Center.
http://modarch.gsfc.nasa.gov/MODIS/LAND/#vegetation-indices, 129P.
Huete, A. R., Liu, H. Q., Batchily, K. and W. J. van Leeuwen (1997) A Comparison of Vegetation Indices Over a Global Set of TM Images for EOS-MODIS. Remote Sensing of Environment 59:440-451.
Huete, A. R., K. Didan, T. Miura, E. P. Rodriguez, X. Gao and G. Ferreira (2002) Overview of the Radiometric and Biophysical Performance of the MODIS Vegetation Indices. Remote Sensing of Environment 83:195-213.
Hughes, G. F. (1968) On the Mean Accuracy of Statistical Pattern Recognizers. IEEE Transactions on Information Theory IT-14: 55-63.
Ishiyama, T., S. Tanaka, K. Uchida, Y. Fujikawa, Y. Yamashita and M. Kato (2001) Relationship among Vegetation Variables and Vegetation Features of Arid Lands Derived from Satellite Data. Advances in Space Research 28(1):183-188.
Jensen, J. R. (1986) Introductory Digital Image Processing. NJ: Prentice Hall. 379p.
Jensen, J. R. (2005) Introductory Digital Image Processing: A Remote Sensing Perspective. 3rded. Upper Saddle River, N.J. : Prentice Hall. 526p.
Jia, X. and J. A. Richards (1994) Efficient Maximum Likelihood Classification for Imaging Spectrometer Data Sets. IEEE Transactions on Geoscience and Remote Sensing 32(2): 274-281.
Kaufman, Y. J. and D. Tanre (1992) Atmospherically Resistant Vegetation Index(ARVI) for EOS-MODIS. IEEE Transaction on Geoscience and Remote Sensing 30(2):261-270.
Lee, J. S. (1980) Digital Image Ehancement and Noise Filtering by Use of Local Statistics. IEEE Transactions on Pattern Analysis and Machine Intelligence PAMI-2(2):165-168.
Lee, J. S. (1983) Digital Image Smoothing and the Sigma Filter. Computer Vision, Graphics and Image processing 24: 255-269.
Lillesand, T. M. and R. W. Kiefer (1994) Remote Sensing and Image Interpretation. 3rded. New York: Wiley. 750 p.
Marceau, D. J., P. J. Howarth, J. M. Dubois and D. J. Gratton (1990) Evaluation of the Grey-level Co-occurrence Matrix Method for Land-cover Classification Using SPOT Imagery. IEEE Transactions on Geoscience and Remote Sensing 28:513-519.
Ready, P. J. and P. A. Wintz (1973) Information Extraction, SNR Improvement and Data Compression in Multispectral Imagery. IEEE Transactions on Communications 35(12):1123-1131.
Rouse, J. W., R. H. Haas, J. A. Schell and D. W. Deering (1974) Monitoring Vegetation Systems in the Great Plains with ERTS. Proceedings, 3ed Earth Resource Technology Satellite(ERTS) Symposium, Vol. 1:48-62.
Running, S. W., C. O. Justice, V. Solomonson, D. Hall, J. Barker, Y. J. Kaufman, A. H. Strahler, A. R. Huete, J. P. Muller, V. Vanderbilt, Z. M. Wan, P. Teillet and D. Carneggie (1994) Terrestrial Remote Sensing Science and Algorithms Planned for EOS/MODIS. International Journal of Remote Sensing 15(17):3587-3620.
Schowengerdt, R. A. (1997) Remote Sensing – Models and Methods for Image Processing. 2rded. New York: Academic Press. 522p.
Shahshahani, B. M. and D. A. Landgrebe (1994) The Effect of Unlabeled Samples in Reducing the Small Sample Size Problem and Mitigating the Hughes Phenomenon. IEEE Transactions on Geoscience and Remote Sensing 32(5):1087-1095.
Swain, P. H. and S. M. Davis (1987) Remote Sensing: the Quantitative Approach. New York:Mcgraw Hill. 180 p.
United Nations (1992) Drafts, Agenda 21, Rio Declaration, Forest principles. United Nations Conference on Environment and Development. Rio de Janeiro, Brazil. A/CONF.151/4(Part Ⅱ). Pp.27-133.
Webster, C. J. (1996) Population and Dwelling Unit Estimates from Space. Third World Planning Review 3:155-176.
Weszka, J. S., C. R. Dyer and A. Rosenfeld (1979) A Comparative Study of Texture Measures for Terrain Classification. IEEE Transactions on System, Man and Cybernetics SMC-6:269-285.
Zhang, Q., J. Wang, P. Gong and P. Shi (2002) Study of Urban Spatial Patterns from SPOT Panchromatic Imagery Using Textural Analysis. International Journal of Remote Sensing 24(21):4137-4160.