本研究主要應用機器視覺技術,完成蝴蝶蘭苗幾何特徵演算法則之建立,應用這些法則所估算蝴蝶蘭在大苗期的幾何特徵値將作爲分級的依據。利用兩部CCD攝影機分別擷取蝴蝶蘭苗前視與上視影像,應用影像處理之技術,包括邊界鏈碼、Hotelling轉換、黃金分割搜尋法、計算葉片數及貝氏分類法等技術建立演算法則,分別估算蝴蝶蘭苗影像之最上層兩葉片端點之距離(葉距)、夾角、葉片長度、葉片長寬比及葉片數等主要幾何特徵。針對44盆蝴蝶蘭進行試驗,應用本研究所建立之演算法則估算幾何特徵,同時以人工方式量測,所估算之結果與實際量測比較,以量測所得之值爲正確値,其中葉片長度之估算最爲準確,其平均相對誤差爲1.80%,而葉片長寬比之平均相對誤差最大,達7.04%,此誤差在實務上仍是可接受的,顯示利用本研究所建立之演算法則可以準確估算蝴蝶蘭主要幾何特徵值。
A methodology using machine vision to estimate the geometric characteristics of phalaenopsis orchid during big plant stage was established in this paper. Those geometric characteristics can be used in sorting process. The images of the phalaenopsis orchid including front and top images were grabbed with two CCD cameras. The image processing techniques including the boundary-chain-codetheorem, Hotelling transformation, the golden section search method, and Bayesian classification were applied to develop algorithms that were used to estimate the geometric characteristics. The characteristics included the distance between two end points, the angle between the midribs, the length, the length/width ratio of the two upper most leaves,and the total number of leaves of each the phalaenopsis orchid during big plant stage.44 samples were investigated. Both estimating by using our method and measuring manually were undertaken. The results were compared. The average relative error between estimated values and measured results of length was1.80% which was the smallest in this study. The average relative error of length/width ratio was 7.04% which was the worst. However, it was acceptable in practice. This showed that the algorithms we developed were capable of estimating characteristic of phalaenopsis quite precisely.