自動化藻類濃縮設備及藻類數位影像辨識之研究

The development of on-line algal concentrating device and phytoplankton counter

郭廷偉(Ting-Wei Kuo)

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(2011年)

https://doi.org/10.6342/NTU.2011.03309

摘要

本研究的目的在於發展一套可自動濃縮藻類的設備，並以自動影像擷取設備進行藻類影像拍攝，最後以辨識軟體對影像上的未知藻類樣本進行辨識以及計數。
研究內容主要可分為三部份。第一部份係浮游生物網藻類自動濃縮裝置的設計與製作，此裝置係利用浮游生物捕集網的原理，結合定時器、抽水馬達以及電磁閥等配件進行組裝。此裝置可藉由定時器控制抽水時間以及電磁閥開關時間以達到調整水樣濃縮倍數的目的。第二部份係影像擷取系統，以國立臺灣大學環境工程學研究所所內水池的藻類作為觀察對象，先以自動濃縮裝置濃縮水樣之後，利用CCD顯微鏡進行藻類影像拍攝。
第三部份係影像處理程式及藻類辨識程式，對水池中常出現的單胞藻、微囊藻以及角鼓藻進行影像擷取、特徵值萃取等步驟建立特徵值資料庫。之後利用貝氏分類法以及最小距離法進行未知物體的辨識。
本研究進行初步藻類影像辨識，結果發現系統對於單胞藻以及角鼓藻有較佳的辨識效率，分別為95.6％、93.3％，對於微囊藻辨識能力較差，辨識率僅有46％。因此進行了特徵值鑑別度的分析，並利用分析結果所得鑑別度較高的特徵值進行第二次的影像辨識。辨識結果為系統對於單胞藻的辨識率為96.4％、對於微囊藻辨識率上升為73.7％、對於角鼓藻辨識率為97.3％，顯示經鑑別度分析後的特徵值確實可提昇系統辨識效益。
從兩次辨識結果發現，微囊藻辨識率從原本46％提昇為73.7％。但對被系統誤判的微囊藻影像進行分析後發現，因微囊藻的形狀不具規則性，所以特徵值資料庫仍稍嫌不足，不能成功辨識各種型態的微囊藻，因此需增加資料庫的數量，以獲取更完善的微囊藻特徵。

關鍵字

微囊藻 ； 影像辨識 ； 藻類濃縮裝置 ； 特徵值鑑別度分析

並列摘要

The aim of this research is to develop an on-line algal cell concentrator, and to capture algal image with automatic image extraction device. Finally these images will be counted and recognized by image recognition and numerating program.
The first part of the research works was designing and manufacturing an automatic algal cell concentrating device, which was composed of a plankton net, a timer, a water pump and electronic valves. Secondly, phytoplanktons taken from the ecological pond at Graduate Institute of Environmental Engineering of National Taiwan University were concentrated by the on-line device. Then, the algal images were acquired by a CCD microscope.
Finally, the images were processed with a phytoplankton recognition system. In order to build the database of features of algal species, including Chlamydomonas, Microcystis and Staurastrum, the algal images were processed with an image processing program, which includes the steps of image segmentation and features extraction. Then those unknown pictures were classified by Baye’s classifier and minimum distance method.
From the results of the feature database establishment and parameter optimization with learning sample sets, the recognition system could better identify Chlamydomonas and Staurastrum with the correct-recognition ratios of 95.6% and 93.3%, respectively. The system showed lower correctness for Microcystis, 46%. By deleting some inefficient features based on the results of discrimination analysis, the correctness ratios for Chlamydomonas, Microcystis and Staurastrum were improved to 96.4%, 73.7% and 97.3%, respectively. The discrimination analysis is able to enhance the performance of the recognition system.
Although the correctness of recognition for Microcystis could be raised to 73.7%, the size of the training database is not big enough to successfully recognize all types of Microcystis due to that the appearance of Microcystis is irregular and complicate. So the size of the training database should be enlarged to acquire more special cases of Microcystis.

並列關鍵字

Microcystis ； image recognition ； algal cell concentrating ； feature discrimination analysis

參考文獻

行政院環保署(2010)，98年版環境白皮書。

謝清祿、鄭聖夫、林達德(1997)，應用影像紋理分析及類神經網路辨識甘藍種苗之生長階段，農業機械學刊。

簡鈺晴(2005)，翡翠水庫藻類多樣性之分析及消長動態之模擬，國立臺灣大學環境工程學研究所，碩士論文。

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

鄭宗欽（2008），自動化影像辨識系統檢測藻類數量及種類方法之研究，國立臺灣大學環境工程學研究所，碩士論文。