航攝數位影像陰影資訊分析及其在森林資源調查之應用

Shadow Information Analysis of Digital Aerial Images and Its Application of Forest Inventory

謝依達(Yi-Ta Hsieh)

指導教授 ： 陳朝圳 ； 鍾玉龍

國立屏東科技大學/農學院/生物資源研究所/博士(2015年)

https://doi.org/10.6346/NPUST.2015.00042

摘要

森林資源調查扮演森林經營管理與監測的重要關鍵，隨著空間資訊獲取方法的日新月異，利用航遙測分析技術於森林資源調查與監測已有具體成效。然而在遙測影像在拍攝過程中，會受太陽角度、地形起伏、地物遮蔽等因素影響而產生陰影，並影響影像品質。事實上，在航遙測光學影像中，陰影一直被視為變遷分析與影像分類等應用層面的影像雜訊，經常導致分析結果之準確度受到影響，因此影像陰影區域處理為被視為一項重要的研究課題。近年來空載多光譜遙測儀器可獲取高空間、高輻射解析力資料，如Leica ADS-40、Intergraph DMC、Vexcel UltraCam-D等多光譜航測影像，皆可獲得12-bit以上的輻射解析力(Digital Number: 0–4,095)，輻射解析力的增加對於陰影區域之解釋提供了極高的潛力。
有鑑於此，本研究的目的如下所述: (1) 以高輻射解析力數位航測影像進行陰影區域的光譜特徵分析; (2)測試三種陰影偵測方法(NIR method, Nagao’s method, Brightness method)，並分析較適用於數位航測影像之方法; (3) 進行陰影區域的土地覆蓋分類，並採用四種陰影補償技術輔助後續分類(方法1，13-bit 陰影光譜資訊; 方法2，線性相關校正陰影補償技術; 方法3，值方圖匹配陰影補償技術; 方法4, 多元資料融合陰影補償技術)，並比較各方法之優劣; (4)利用陰影比例技術結合高解析力數位航測影像進行林分性態值推估，並討論影響林分性態值推估的影響因子(陰影比例校正樹冠陰影標準化、取樣大小、林型差異)。
研究結果指出，(1)陰影光譜的特徵可歸納下列幾點 (i)陰影區的像元值比非陰影區域低很多; (ii)陰影區的光譜資訊會受到不同地物所影響，代表陰影區的光譜具有恢復到非陰影區域的潛力; (iii) 由於散射效應的影響，可見光的波長越長，陰影區的像元值會越低; (iv) 植生在陰影區域時亦會強烈反射近紅外光; (v) 一般常用植生指數(NDVI, SAVI)在陰影區域亦具有區分植生與非植生的能力。(2)本研究測試三種陰影偵測方法，結果指出Brightness method 和Nagao’s method 明顯較NIR method為好。(3) 在進行陰影區域的土地覆蓋分類的結果中，其中方法1(13-bit 陰影光譜資訊)與方法2(線性相關校正陰影補償) 所得結果較佳(90%以上的總體精確度)，，成果證明13-bit的高輻射解析力航攝數位影像具有足夠資訊具有勝任陰影區域土地覆蓋分類的潛力。(4) 利用陰影比例技術結合高解析力數位航測影像進行林分性態值推估的結果指出，(i)全區推估的結果呈現顯著的線性關係但較低的R_adj^2值; (ii) Leboeuf et al. (2007)所提出的陰影比例校正法對於本研究的推估沒有明顯幫助; (iii) 地面調查樣區的樣區大小推估成果較好(iviii) 以陰影比例將各林型分開進行推估明顯提升推估成果在不同林型的林分性態值推估結果中指出，混淆的林型推估效果較差，較同質性的林型推估效果較好。
綜合上述結果可知，採用高輻射解析力航攝數位影像具有解決陰影問題的潛力，以及整合陰影遙測資訊可提升森林資源調查技術之應用。

關鍵字

高輻射解析力 ； 土地覆蓋分類 ； 陰影光譜 ； 陰影偵測 ； 陰影補償 ； 陰影比例 ； 林分性態值

並列摘要

Forest resource inventory plays a key role in forest management and monitoring. Using photogrammetry and remote sensing techniques in forest resource inventory has obtained significant results such us as getting the spatial information of forest. However, during the image capturing process, numerous influential factors hinder the quality of these images, such as the shadows caused by the different angle of the sun, terrain features, and surface object occlusion. Furthermore, the shadows in optical remote sensing images are regarded as image nuisances in numerous applications, specifically, change detection and image classification frequently affecting the accuracy of analytical results. Therefore, research on shadow processing in images is highly valued. In recent years, airborne multispectral aerial image devices have been developed high radiometric resolution data, including Leica ADS-40, Intergraph DMC. These devices are capable of capturing radiometric resolution images of 12 bits or higher, for example, a 12-bit digital number (DN) ranged from 0–4,095. The increased radiometric resolution of digital imagery provides more radiometric detail of potential use in classification or interpretation of land cover of shadow areas.
Therefore, the objectives of this thesis are as follows (1) investigating the spectral properties of shadow areas by high radiometric resolution images; (2) testing three different shadow detection methods (NIR method, Nagao’s method, Brightness method), and look for a suitable shadow detection method from digital aerial images; (3) classification of the shadow areas was tested by using four compensation methods (Method 1, used 13-bit spectral information in shaded areashadow areas for classification; Method 2 used Linear Correlation Correction (LCC) before the classification; Method 3 used Histogram Matching (HM) before the classification; and Method 4 used MultiSource Data Fusion (MSDF) to aid in classification of shadows.), and compared the benefits of those shadow compensation methods; (4) estimating stand attributes using the shadow fraction method and high spatial resolution digital aerial images, also discuss the impacts of estimation of stand attribute due to three two factors (tree shadow normalization, different sample size, forest type).
The results indicated that (1) the spectral characteristics of the shadow area can generalize that (i) the DN values in shadow area are much lower than in nonshadow area; (ii) DN values received from shadowed areas that can be also affected by different land cover, and showing the possibility of land cover property retrieval as in nonshadow area; (iii) the DN values received from shadowed regions decreases in the visible light from short to long wavelengths due to scattering; (iv) the shaded areashadow area NIR of vegetation category also shows a strong reflection; (v) vegetation indexes (NDVI, SAVI) still have utility to classify the vegetation and non-vegetation in the shaded areashadow area. (2)Three shadow detection methods were tested in this study, and the results indicated Brightness method and Nagao’s method are significantly better than NIR method. (3) Method 1 (13-bit high radiometric resolution images; no treatment) and method 2 (LCC) presented better land cover classification results and possessed significant accuracy (over 90% of overall accuracy), and this result further demonstrates that the 13 bit high radiometric resolution images (ADS-40) have sufficient information to satisfy classification of shadow areas.and this result further demonstrates that the spectral data of 13 bit high radiometric resolution images (ADS-40) have potential for the classification of shadow areas. (4) the results of estimation of stand attributes using the shadow fraction method and high spatial resolution images show that (i) global estimation results show linear relationships but have low values for the R_adj^2 ; (ii) the tree-shadow normalization procedure proposed by Leboeuf et al. (2007) is not significantly beneficial for estimating stand attribute in our dataset; (iii) The result indicates that inventory plot size got the highest R_adj^2; (iviii) Our the results of resultsthe stand attribute estimation of different forest type clearly displayed show that estimation of stand attributes for each forest type separately improved the results.mixed forest type can affects the accuracies of estimations, and higher homogeneity forest type could obtained better estimation.
From above results, the high radiometric resolution digital aerial images show potential to deal with shadow problem and integrate shadow information to improve the forest inventory techniques.

並列關鍵字

High radiometric resolution ； Land cover classification ； Shadow spectra ； Shadow detection ； Shadow compensation ； Shadow fraction ； Stand attribute

參考文獻

Hsieh, Y. T., Y. L. Chung, C. S. Liao, Y. G. Yui, K. C. Teng, and S. T. Wu (2011)b Application On High Resolution Digital Aerial Images for Automatic Forest Fires Mapping by Change Detection Techniques.(in Chinese). Journal of Photogrammetry and Remote Sensing 16(1): 11-22.

Lei, T. C., C. Y. Li, H. C. Yeh, and S. Wan (2009) The Study on Regional Object Classification for Paddy Rice Patch Extraction from ADS-40 Image. (in Chinese) Journal of Photogrammetry and Remote Sensing 14(2): 127-140.

Yeh, K. S., Li, M. Y., and Wang, Y. H. (2010) Applying digital aerial images in natural hazards mitigation in Taiwan. (in chinese). Journal of Photogrammetry and Remote Sensing 15(1): 123-140.

Adeline, K. R. M., M. Chen, X. Briottet, S. K. Pang, and N. Paparoditis (2013) Shadow detection in very high spatial resolution aerial images: A comparative study. ISPRS Journal of Photogrammetry and Remote Sensing 80: 21-38.

Asner, G. P., and A. S. Warner (2003) Canopy shadow in IKONOS satellite observations of tropical forest and savannas. Remote Sensing of Environment 87: 521-533.