自動化影像辨識系統檢測藻類數量及種類方法之研究

Automatic recognition and numerating of phytoplankton with microscopic imaging

鄭宗欽(Cheng, Tsung-chin)

指導教授 ： 吳先琪

國立臺灣大學/工學院/環境工程學研究所/碩士(2008年)

https://doi.org/10.6342/NTU.2008.02763

摘要

本研究結合流動式進樣裝置、數位影像擷取、數位影像處理以及藻類辨識，建立一套可應用於現場的浮游藻類自動監測系統。第一部分為進樣裝置之設計，係利用自製的流動式樣品槽結合間歇性泵浦，使水樣能夠自動進樣至顯微鏡聚焦處。第二部分為影像擷取系統，利用CCD數位相機拍攝藻類影像，以台大醉月湖之水樣做為觀察對象，得到待辨識之藻類影像。第三部分為影像辨識及計數系統，係就柵藻、鐮形纖維藻、藍綠藻以及盤星藻，利用各藻類間特徵值的差異在貝氏分類器下進行辨識；藻類計數方面，利用大量的擷取影像，以影像中總藻類數量與水樣總體積對水體的藻類數量進行估計。

此系統的藻類定性片辨識率盤星藻有辨識率63.9%，藍綠藻61.6%，鐮形纖毛藻61.1%，而柵藻辨識率51.6%。使用間歇流動攝影方面，因為盤星藻與雜質易交雜在一起使藻類影像複雜，以及盤星藻影像並非皆正面朝向鏡頭而使得辨識率不佳，僅有14.3%。柵藻、藍綠藻之辨識率分別為56.3% 及76.2%，而鐮形纖維藻由於季節的影響於醉月湖中並未見到。

辨識系統測試的結果辨識率雖不盡理想，此系統如持續改進，應可以發展為自動監測水中浮游藻類之有力工具。

關鍵字

藻類 ； 影像辨識 ； 數位影像處理

並列摘要

The aim of this research is to develop a system for automatic monitoring phytoplankton in real water bodies. The system includes sample injector, digital image capture system , digital image process, and pattern recognition. The first part of the system is the sample injector including the flowing cell the pump pumping sample to the focal point of the microscope intermittently. The second part of the system is image capture system in which CCD camera captures images and the acquired images were processed for pattern recognition based on the sample from Zui-Yue Lake in National Taiwan University. The third part of the system is computerized pattern. The Bayes'classifier processes were used for the pattern recognition of the algae features. The system is able to estimate the numbers of phytoplankton by total water volume and the numbers of phytoplankton in the plenty of captured images.

The recognition accuracy of the algae specimen achieves 63.9%; 61.6% for Chroomonas; 61.1% for Ankistrodesmus; 51.6% for Scenedesmus. In intermittent flowing and capturing images, The recognition accuracy is down to 14.3% for Pediastum because Pediastum are tend to mix with other objects and rotating quite often; the recognition accuracy achieves 56.3% for Scenedesmus; 76.2% for Chroomonas. Ankistrodesmus were not observed due to the season.

Recognition accuracy of the recognition system is not good enough. With further improvement the system has the potential to be a powerful automatic monitoring device
for phytoplankton in the future.

並列關鍵字

phytoplankton ； pattern recognition ； digital image process

參考文獻

蔡郁佳（2007），浮游藻類之種類數量檢測及辨識系統之研究，國立台灣大學環境工程學研究所，碩士論文。

蘇郁婷（2006），微囊藻毒素在淨水處理程序流佈之研究，國立成功大學環境工程 學系，碩士論文。

Buskey, E. J. and Hyatt C. J. (2006) Use of the FlowCAM for semi-automated recognition and enumeration of red tide cells (Karenia brevis) in natural plankton samples. Harmful Algae 5(6): 685-692.

Givan, A. L. (2001). Flow Cytometry: First Principles 2nd Ed. Wiley-Liss, New York.

Hu, M. K. (1962) Visual Pattern Recognition by Moments in Image Processing. IRE Trans. Info. Theory, 8: 179-187.