浮游藻類之種類數量檢測及辨識系統之研究

Abstract

本研究結合流式細胞技術、數位影像擷取、數位影像處理以及樣 式辨認，企圖建立一個監測水中浮游藻類的系統。本研究共分為四大 部分，第一部份探討系統流動式樣品槽之設計與製作，接著結合顯微 鏡與數位相機在樣品槽中拍攝藻類影像，第三部份則為影像前處理， 目的是將第二部份之影像能夠輸入第四部份的樣式辨認。 利用手工製作的流動式樣品進樣槽，結合程控之蠕動泵浦，將含 有浮游藻類的水樣驅動流過樣品槽，並將顯微鏡聚焦至樣品流中，再 利用數位相機透過顯微鏡拍攝藻類的影像，以台大環工館水池之水樣 為例，可得到做為後續樣式辨認用之藻類影像。 第三部份建立之影像前處理程序，目前已可處理人工培養之藻類 溶液之影像，在第四部份之樣式辨認。最後樣式辨認的部份則測試了 現場採樣之藻類定性片的影像與人工培養藻類於樣品槽中之影像，結 果發現基於單一高斯密度函數之貝氏分類器對於藻類定性片的影像有 不錯的辨識率，除了對於有方向性之藻類效果不佳，例如角股藻僅有 40％ 之辨識率其餘均大於80％。而在人工培養的藻類影像方面，由於 人工培養之藻類彼此間外型差異不大，測試之結果並非十分理想辨識 率的範圍在64％到100％間。初步之研究顯示此系統如繼續改進，應可 發展為自動監測水中浮游藻類之有力工具。

The aim of this research is to develop a system for monitoring the types and abundance of phytoplanktons in water. Flow cytometry, digital image capturing, digital image processing, and pattern recognition(PR) were integrated in the system. Four major sets of researches were performed in this study. The first set of the research workes were the design and fabrication of the flowing cell. This was followed by image capturing by digital camera through microscope. The third set of research was image pre-processing, then the images obtained from the secong step were processed with developed PR technique in the fourth step. By pumping the algal suspension sampled from the garden pond in my institute into the handcrafted flowing cell with a programmed metering pump, focusing the microscope to the cell chamber and taking picture with a digital camera, images of phytoplankton with satisfactory resolution for the following pattern recognition could be captured. The developed image pre-processing procedure is able to select the images of algae from laboratory cultures and to save them for the later pattern recognition procedure. The images from either the algal specimen made from field sample or laboratory cultured algae flowing through the flow cell were used for the pattern-recognition tests. The results show that the Bayes’ classifier based on the single Gaussian probability density function is able to classify images of four types of algae from specimen with 88% to 100% successful recognition, except for the algae with specific orientational characteristics, for example, Staurastrum sp. Due to less difference among the laboratory cultured species, the successful recognition of them ranges from 64% to 100%. The preliminary study indicates that with further improvement the integrated system has the potential to be a powerful automatic monitoring device for phytoplankton in the future.