本研究旨在測試不同植物生長指數並建立紅藻之影像辨識方法。在桃園藻礁分佈著許多紅藻門(Rhodophyta)的藻類，其中主要的造礁藻類為無節殼狀珊瑚藻(crustose coralline algae, CCA)。目前紅藻的調查方式，受限於取樣方式以及現場環境因素，較難達成效率以及標準化調查，因此本研究希望利用可見光無人機空拍影像來輔助傳統紅藻之調查。本研究於大潭藻礁G2區及觀新藻礁保護區保生樣站兩個樣站進行空拍作業取得影像，並於實驗室利用地理資訊系統ArcGIS以人工目視法的曲線繪製功能標示出包括紅藻、礁體上之空洞等目標物。本研究利用MATLAB讀取原始影像的紅色(R)、綠色(G)、藍色(B)三者可見光波段之DN數值，正規化後轉換為色度座標，利用不同顏色指標組合、植物生長指數等因素統整歸納，嘗試找出適合做為自動辨識之閾值，以達到紅藻的自動辨識，並進行紅藻的覆蓋率計算。本研究測試了11種包含了顏色指標、顏色指標組合以及植物生長指數，結果顯示以過剩綠指數(Excess Green)可以最有效辨識出紅藻之位置；此指數可以有效排除非紅藻與影像過曝之高亮度區域；經測試後，過剩綠指數<-0.032為較適當的自動化辨識之閾值。本研究利用混淆矩陣計算其可信度，所得的平均Kappa 值為0.5244，平均整體正確度(Overall Accuracy, OA)為87.79%。一般認為Kappa 值介於0.4～0.8 之間具有中等的信賴度，因此本研究的自動辨識結果應具有一定的信賴度，可見光無人機空拍影像可以做為傳統紅藻調查之輔助，提高調查作業效率，並計算其覆蓋率。;This study aims to test different vegetation indices and establish an image recognition method for red algae. In the Taoyuan algae reef, many algae of the Rhodophyta phylum are present, with the primary reef-building algae being crustose coralline algae (CCA). Current red algae investigation methods are limited by sampling techniques and field environmental factors, making efficient and standardized surveys challenging. This study uses visible light drone aerial imagery to assist traditional red algae surveys. Aerial photography was conducted at two sites: G2 area of Datan algae reef and Baosheng sampling station in Guanyin algae reef protection area. Images were manually analyzed in ArcGIS to delineate targets such as red algae and holes on the reef. MATLAB was used to read the red (R), green (G), and blue (B) visible light band values of the images, identify their spectral reflection characteristics, and integrate different band combinations, vegetation indices, and habitat information to determine suitable thresholds for automatic recognition. The study tested 11 vegetation indices, finding that the Excess Green Index most effectively identified red algae, excluding non-red algae and high-brightness areas from overexposure. The optimal threshold for automated recognition was Excess Green Index <-0.032. Reliability was calculated using a confusion matrix, resulting in an average Kappa coefficient of 0.5244 and an average overall accuracy of 87.79%. A Kappa value between 0.4 and 0.8 indicates moderate reliability. Thus, visible light drone aerial imagery can effectively assist traditional red algae surveys, improving efficiency and coverage rate calculations.
