| 摘要: | 本計畫分六個工作項目，包括 (1) 光達高程過濾與森林材積計算， (2) 高解析影像與光達資料整合於三維森林模型之建立與變遷偵測， (3) 高解析遙測影像於特定林區植被變遷之監測， (4) 異質遙測影像分類技術用於林地分布之監測， (5) 利用全偏極干涉雷達估算材積， (6) 應用MODIS 衛星資料在台灣地區森林資源監測之應用。各項目地工作內容摘要如下：項目一：光達高程過濾與森林材積計算：本研究前期已完成空載光達之資料收集、初步精度驗證、及誤差分析研究等。空載光達所測得之數據須經由高程過濾與編修之作業以分離覆蓋面與地形面，研究顯示這種過濾作業人工佔60-80%。國際航測遙測學會研究顯示都會區與地形複雜地區是最困難地區，而且點雲密度對過濾效果的影響尚無任何瞭解。本研究之目的在以實務考量下，探討高程過濾之自動與人為作業，並進而區分地面點與非地面點如生物覆蓋面與人造物覆蓋面等。此成果可提供後續根據兩數據面之差，作為估算材積與生物量等應用。項目二：高解析影像與光達資料整合於三維森林模型之建立：本研究目標為建立一個實用的系統以作為林業管理所需之森林三維覆蓋模型之工具。以融合富含光譜之高解析影像及形狀資訊的光達出發重要工作項目及實施方法包括 (1) 自動化影像與光達點雲之套合， (2) 數位影像之校正， (3) 森林區偵測， (4) 樹林覆蓋模型之建立，及 (5) 可靠度與精度分析。項目三：高解析遙測影像於特定林區植被變遷之監測：本研究內容包含技術性及服務性的工作。技術性的工作基本上是以高解析遙測影像(SPOT-5或福衛二號)為偵測工具，針對特定區域之國有林地進行變遷監測，開發包含影像自動化的變遷偵測，及與圖資相疊合的方法。服務性的工作主要是持續維護多譜掃瞄資料庫，並舉辦一次國內空載及衛載遙測影像之應用近況及成果展示座談會。項目四：異質遙測影像分類技術用於林地分布之監測：光學影像已經廣泛應用在土地覆蓋與土地利用分類上，但僅使用單一感測器會限制分類的精度，特別是在亞熱帶的台灣，森林地區經常是較破碎的，且樹種也是較複雜的。因此，開發森林及樹種分類技術乃為當前重要的課題。本計畫之目的即為開發異質影像數據融合(如光學與雷達融合)技術來有效整合多元遙測影像，以提升林地樹種分塊分類的精度。項目五：利用全偏極干涉雷達估算材積：全偏極合成孔徑雷達能提供各類目標物對於不同偏極雷達波反應差異之特性，各偏極雷達波所反應的特性為目標物之介電特性及幾何結構，因此能進行精細之目標物分類, 相異於光學影像之光譜特性。由於植物之幹徑、葉形大小、生長結構、含水量等均會影響全偏極雷達之反應，而這些差異正是分類最佳的特徵。我們利用多時多偏極的ENVISAT-1雷達影像，分析植生量與雷達回波量之關係並進一步發展植生量反演程序與驗證反衍精度。項目六：應用MODIS衛星資料在台灣地區森林資源監測之應用：本研究之主要目標為應用MODIS、ROCSAT-2衛星資料來遙測台灣地區的森林資訊，由於MODIS資料具有大範圍涵蓋、較多光譜頻道的優點，經由其優越的頻道輻射特性，求取得更精確的植被指數值及大氣效應，以得到更準確的森林資訊，另外也藉由其光譜、空間解析的優點，建立雲、林煙、林火的分類與標定，有效地監測森林的變化。另外本研究也運用ROCSAT-2高解析的優勢，分析小區域植被資訊，並檢驗其與其他遙測影像在輻射融合方面的應用。 This project comprise six items of investigations, namely (1) Lidar DEM filtering and application on wood volume estimation, (2) Fusion of High Resolution Imagery and Lidar Data for the Reconstruction of 3-D Forest Canopy, (3) Change Detection of Forest Land Using High Spatial Resolution Remotely Sensed Images, (4) Image Classifiers Using Heterogeneous Remote-Sensing Data for Monitoring the Distribution of Forest Areas, (5) Estimating of timber volume using fully polarization SAR, and (6) Applications of MODIS satellite Data in Forest Resources Monitoring in Taiwan Area. The major works are summarized as follows. Item 1: "Lidar DEM filtering and application on wood volume estimation": So far, our study on LiDAR has covered the data collection on pilot areas, accuracy assessment, and error analysis. For obtaining the elevation of bare earth surface from LiDAR data, human interaction consumes some 60-80% of processing time. The report of ISPRS WG-III3 shows that errors due to imperfect of filters are mostly in complex urban areas and rough terrain with vegetation, which is the case in Taiwan. And, the influence of point density could not well be determined in ISPRS experiment. Thus, in this study, data acquired in forest area of mountainous terrain in Taiwan will be processed by selected software with algorithms and parameters assessed for optimal performance. Item 2: "Fusion of High Resolution Imagery and Lidar Data for the Reconstruction of 3-D Forest Canopy": The objective of this investigation is to establish an applicable method that reconstructs 3-D forest canopy model for the management of forest. Major works include (1) registration of images and Lidar point clouds, (2) geometric correction for the images, (3) detection of forest blocks, (4) establishment of the canopy model, and (5) analyses of reliability and accuracy. Item 3: "Change Detection of Forest Land Using High Spatial Resolution Remotely Sensed Images": The contents of the project include practical and serviced tasks. The practical task will use the high spatial resolution satellite images (SPOT 5 or ROCSAT 2) to perform the change detection for forest land. The automatic change detection algorithm and the map overlaid technique will also be developed in the project. The serviced task will continuously maintains the database of airborne multi-spectral scanner data, and organize a workshop for the project reports. Item 4: "Image Classifiers Using Heterogeneous Remote-Sensing Data for Monitoring the Distribution of Forest Areas": Optical images are widely used in land cover and land use classification. However, the use of single sensor data often limits the accuracy of work, especially for the subtropical Taiwan forest area where fragmentation is a common phenomenon and the types of the trees are often complicated. Therefore, developping a suitable technique to classify forests and trees is an important task. In this study, multi-sensor image fusion (optical and radar) technique will be developed to enhance the segmentation and classification accuracy. Item 5: "Estimating of timber volume using fully polarization SAR": Fully polarimetry SAR provides a thorough description of the scattering properties of radar target materials. The dielectric and geometric properties, including trunk, leave size, structure, and moisture content, response differently to different polarizations and thus offer the opportunity to differentiate forest structure from polarimetric response that the optical sensors can not provide. In this project, the radar vegetation index using the multi-temporal and multi-polarization ENVISAT-1 SAR signatures will be investigated. A retrieval algorithm will be developed to estimate th 研究期間：9401 ~ 9412 |
