多重解析度影像應用於水稻坵塊萃取適用性研究

Abstract

如何更快速、更準確地由航遙測資料中獲取地理訊息，為近年來熱門的研究議題。傳統衛照影像與航照影像受限於影像解析度，在臺灣農業土地多樣性極高的情形下，分類之精確程度可能有其限制。但近年來發展出以無人飛行載具（UAV）搭載非傳統航測專用攝影機所組成的無人飛行系統（UAS），其快速起降、低空作業、高解析度...等特性，對於農業作物不同生長時期影像紋理之辨識應有極高應用價值。本次實驗在以快速進行影像分析、盡可能減少人工判釋的前提下，分別對衛照影像、航照影像、UAS小像幅影像，引用物件導向式概念進行物件分割後對水稻進行分類，最後評估萃取坵塊精確度。實驗結果顯示衛照影像受限於地元解析度，無法於物件分割時獲得與現況較為相符的成果，進而影響分類精確度。航照影像分類精確度與UAS小像幅影像相比雖略有不足，但整體而言，應用於特定作物萃取已算是相當足夠。若將研究區鄉鎮全區以本實驗方式進行半自動化分類，與傳統外業踏勘相比，可節省不少工作人天。由實驗結果證實，UAS小像幅影像應用物件導向式分類於半自動化分類判釋農作，綜觀其作業效率與精確度，確實具有相當潛力。How to faster and more accurate extract geospatial information from the remotely sensed data has been a popular research topic recently. Traditional satellite image and aerial image are limited to image resolution. In Taiwan, due to the agricultural land diversity, accuracies from traditional classification techniques have its limits. For the past few years, the development of Unmanned Aerial System（UAS）, which is composed of Unmanned Aerial Vehicle（UAV）and off-shelf commercial camera has attracted interests in many geo-information related fields. With capabilities of fast take-off and landing, working at much lower altitude, and much higherspatial resolution, UAS has showed high potential in image acquisition and interpretation of crops at different growth stages. On the premise of fast image analysis and reducing manual interpretation, we applied the concept of object-oriented classification to extract objects from satellite, aerial and UAS image respectively. Then, we classify rice paddies and evaluate the accuracy of the extracted paddies. The result shows that satellite image is limited by its ground sampling distance, and further affected in classification accuracy. Comparing with UAS image, the accuracy of aerial image classification is slightly lower, but still good enough for specific crop paddies extraction. Comparing with traditional field survey, the semi-automatic classification procedure proposed here can save a lot of working days and man power in the field. Experimental results confirm that by applying object-oriented technique UAS image has very high potential in semi-automatic crops paddies extraction.