近年來農村的勞動力流失已成為國內產業結構的嚴重問題。而糧食的自給率影響到一個國家的經濟結構穩定性。過去十年，在全球氣候快速變遷導致農作欠收的現況下，我國僅剩下三成的糧食自給率顯得岌岌可危。基於預防勝於治療的道理，積極的培養國內的農業發展並且提倡農業自動化的升級，透過農業生長監控的自動化等先進的農耕技術，都可以有效提升農業的產量，進而促進國內農業就業環境的健康發展。準確的測量生物量不僅有助於統計長期的收穫量，還能從生物量的評估中了解到如何調整農地的經營策略，如植物的含水量與灌溉的次數是否達到平衡、施肥肥力及除蟲除草藥劑劑量是否足夠。
本研究所提出的生物量自動監測系統主要是透過training sample extraction、image segmentation、noise removing、image stitching與growth rate estimation五個步驟來偵測覆蓋率。首先系統偵測影像中的混亂程度，接著利用SLIC (simple linear iterative cluster)演算法建立超像素，並利用混亂程度計算出變異閥值以篩選內部像點一致性高的超像素作為訓練集，接著使用K-means演算法將影像分割為茶樹與土壤兩類，再使用SLIC (simple linear iterative cluster)演算法建構更密集的超像素以投票的方式進行影像去雜訊，最後利用GoPro公司開發的Kolor Autopano Giga 3.7影像縫合軟體將茶園內不同區域的分割結果縫合成完整的茶園分割結果，並且經過比對不同時間點之茶園分割結果來估計茶園的成長率。
本研究不同於常見的衛星影像與航拍影像監測，改善了過去監測方式的缺點，使用無人飛行載具之空拍影像監測茶樹覆蓋率，得以少量的人力進行大範圍的監控。另外，針對不同的影像複雜程度使用的分割參數也不相同，自動的選擇分割參數讓系統在測量生物量時能更加準確。本研究設計了不同分割參數正確率分析、與自動選擇參數效率分析的實驗。實驗結果證實本系統可實現自動提供準確之良好分割參數且提供可靠且正確的生物量估計結果。在未來，長期記錄農業收穫量並建立常模，可以幫助耕作者與經營者快速判農地生長的情形，以便實施正確的農地經營策略。

In recent years, the loss of rural labor has become a serious problem of domestic industrial structure. The self-sufficiency rate of food affect the structural stability of a country's economy. based on Prevention is better than treatment, actively cultivate and promote the development of domestic agricultural agriculture automated upgrade is very important. agricultural growth through automated monitoring and other advanced farming techniques, can effectively enhance the promotion of agricultural production, thus contributing to domestic agricultural employment environment healthy growth.
The system can be divided into five steps: training sample extraction, image segmentation, noise removing, image stitching and growth rate estimation. First, the system measure the impurity degree in the image, and then use SLIC (simple linear iterative cluster) algorithms to build superpixel, and using the impurity degree of confusion threshold to filter the internal consistency of superpixel as the training set, then using the K-means algorithm to segment tea and soil. And then use SLIC (simple linear iterative cluster) algorithm to construct more intensive super pixels to denoising by voting. Finally, system using GoPro developed image stitching software Kolor Autopano Giga 3.7 stitching segmentation results of different regions within the tea into a complete tea segmentation results, and after match tea at different points of time division results to estimate the growth rate of tea.
In this study, different from the common satellite image and aerial surveillance images, monitoring methods to improve the shortcomings of the past, system using unmanned aerial vehicle to shot images of tea garden. In addition, each image has different impurity degree. system need to using different parameters to segment image. system automatically selects parameters when measuring biomass can be more accurate. This study was designed experiment to analyze different partition parameter accuracy and efficiency of automatic selection parameter analysis of experiments. The experimental results show the system can automatically provide accurate segmentation parameters of good and provides reliable and accurate estimates of biomass. In the future, long-term record agricultural harvest and the establishment of norms that can help farmer and agriculturists to find the best policy.

[1] "Council of agriculture, Executive Yuan" [Online]. Available: http:// agrstat.coa.gov.tw/sdweb/public/common/CommonStatistics.aspx.
[2] "The international labour organization (ILO)," [Online]. Available: http:// www.ilo.org/global/industries-and-sectors/agriculture-plantations-other-rural-sectors/lang--en/index.htm.
[3] "General of budget, accounting & statistics (DGBAS)," [Online]. Available: http://www.stat.gov.tw/ct.asp?xItem=39451&ctNode=516&mp=4.
[4] Vin, "Forest mensuration and practice," 2004.
[5] N. T. U. S. Laboratory, [Online]. Available:http://seed.agron.ntu.edu.tw/.
[6] P. West, "Tree and forest measurement," Berlin: Springer-Verlag, 2004.
[7] Z. Somogyi and E. Cienciala, "Indirect methods of large scale," European Journal of Forest Research, vol. 126, pp. 197-207, 2007.
[8] S. Gower, C. Kucharik, and J. Norman, "Direct and indirect estimation of leaf area index, fAPAR and net primary production of terrestrial ecosystems," Remote Sensing of Environment, vol. 70, pp. 29-51, 1999.
[9] S. Salis, M. Assis, and P. Mattos, "Estimating the aboveground biomass and wood volume of savanna woodlands in Brazil's Pantanal wetlands based on allometric correlations," Forest Ecology and Managemen, vol. 228, pp. 61- 68, 2006.
[10] R. Pilli, T. Anfodillo, and M. Carrer, "Towards a functional and simplified allometry for estimating forest biomass," Forest Ecology and Management, vol. 237, pp. 583-593, 2006.
[11] P. Muukkonen, "Generalized allometric volume and biomass equations or some tree species in Europe," European Journal of Forest Research, vol. 126, pp. 157-166, 2007.
[12] G. West, J. Brown, and B. Enquist, "A general model for the origin of allometric scaling laws in biology," Science, vol. 276, pp. 122-126, 1997.
[13] S. Brown and A. Lugo, "Biomass of tropical forests: a new estimate based on forest volumes," Science, vol. 233, pp. 1290-1293, 1984.
[14] R. Whittaker and G. Likens, "Methods of assessing terrestria productivity," New York: Springer-Verlag, pp. 305-328, 1975.
[15] S. Brown and A. Lugo, "Biomass of tropical forests: a new estimate based on forest volumes," Science, vol. 233, pp. 1290-1293, 1984.
[16] J. Fang, G. Wang, and G. Liu, "Forest biomass of China: an estimate based on the biomass-volume relationship," Ecological Applications, vol. 8, pp. 1084-1091, 1998.
[17] Y. Pan and T. Luo, "New estimates of carbon storage and sequestration in China's forests: effects of age-class and method on inventory-based carbon estimation," Climatic Change, vol. 67, pp. 211-236, 2004.
[18] G. Zhou, Y. Wang, and Y. Jiang, "Estimating biomass and net primary production from forest inventory data: a case study of China's Larix forests," Forest Ecology and Management, vol. 169, pp. 149-157, 2002.
[19] J. Smith, L. Heath, and J. Jenkins, "Forest volume-to-biomass models and estimates of mass for live and standing dead trees of U. S. forests," pp. 1-62, 2003.
[20] C. Huang and J. Zhang, " Spatio-temporal changes of forest carbon reserves in Sichuan province," Applied Ecology, vol. 18, pp. 2687-2692, 2007.
[21] M. Zhao and G. Zhou, "Estimation of biomass and net primary productivity of major planted forests in China based on forest inventory data," Forest Ecology and Management, vol. 207, pp. 295-313, 2005.
[22] "Guidelines for national greenhouse gas inventory," IPCC, [Online]. Available: http://www.ipcc-nggip.iges.or.jp/.
[23] D. Turner, G. Koepper, and M. Harmon, "A carbon budget fo forests of the conterminous United States," Ecological Applications, vol. 5, pp. 421-436, 1995.
[24] H. LÊewe, G. Seufert, and F. Raes, "Comparison of methods used within Member States for estimating CO2 UNFCCC and EU Monitoring Mechanism: forest and other wooded land," Biotechnologie Agronomie Societe et Environnement, vol. 4, pp. 315-319, 2000.
[25] J. Fang, A. Chen, and C. Peng, "Changes in forest biomass carbon storage in China between 1949 and 1998," Science, vol. 291, pp. 2320-2322, 2001.
[26] J. Canny, "A computational approach to edge detection," IEEE transactions Pattern Analysis and Machine Intelligence, pp. 679-714, 1986.
[27] M. Sezgin and B. Sankur, "Survey over image thresholding techniques and quantitative performance evaluation," Electronic Imaging, vol. 13, pp. 146-165, 2004.
[28] M. Cohen, S. Elder, C. Musco, C. Musco, and M. Persu, "Dimensionality reduction for k-means clustering and low rank approximation," ArXiv, 2014.
[29] A. Radhakrishna, S. Appu, S. Kevin, L. Aurelien, F. Pascal, and S. Sabine, "SLIC superpixels compared to state-of-the-art superpixel methods," Latex Class Files, vol. 6, pp. 1-8, 2011.
[30] "Wiki HSL and HSV," [Online]. Available: https://en.wikipedia.org/wiki/ HSL_and_HSV.
[31] "Kolor autopano giga 3.7, " GoPro, [Online]. Available: http://www.kolor. com/2014/10/20/panorama-software-autopano-pro-giga-3-7-final/.