Research On Diatom Algae Identification Methods Based On Machine Vision

Algae are a very easy plant, commonly located in the freshwater and ocean, fall intoprokaryotic algae and eukaryotic algae, or by categories divided into11gates. Diatoms belong toone of the more important categories of algae. Algae play an important role in environmentalmonitoring, biological pharmaceuticals and other aspects. But traditional classification of algae,for the operator’s requirements is high. Operators need to have deep knowledge of algae formaccumulation, but the intensity of work is large, strong and subjective, and error-prone. In thispaper, using the digital image processing and pattern recognition knowledge for Diatom algaeimage automatic recognition research. The thesis mainly works as follows:Firstly, the significance of the subject and the research status at home and abroad is described,and how the diatom classification do in traditional way and its defections is shown, and theexisting problems of using the image processing technology to achieve the classification ofdiatom in recent years is also described.Secondly, the acquisition of ROI is described in detail. The first step is to get the edge imageby using Sobel edge detection. The second step is to get the binary image by using Otsusegmentation. The third step is to get the contour image by using contour following and floodfills. The fourth step is to remove the noise and vernier by using a small area filter. The fifth stepis to get the circumcircle of object by using external moment contraction method.Thirdly, the classification of diatom cells based on target architecture is proposed.Circumscribed circle distancely divided into10feature extraction rings, and the diatom targetfeature vector is extracted by applying the measures of target structural elements, measures oftarget structural change, measures of target structural angle.Finally, Euclidean distance is uesd to measure diatom image similarity, and the diatom cellimages are classified by using a comprehensive similar distances. Experimental Gallery has102species of306pictures. The evaluation methods of this thesis differs from the evaluation ofclassifiers, and this thesis use a content-based image retrieval commonly used indicators-precision and recall rates. The experimental results of the automatic search comparison show thatthe proposed method of diatom image classification has average accuracy rate of5.4%with afull value of6%and the average recall rate of90.0%with a full value of100%. The proposed method has been compared to similar methods and got better identification and classificationresults.