Automatic Identification of Diatoms with Circular Shape using Texture Analysis

Diatoms are unicellular microscopic algae found in practically any moist environment. Identification of diatom has application in many disciplines, including ecology, archaeology and forensic science. In recent years, the work has been undertaken for automatic identification of diatom. However, the diatom with circular shape has not yet been considered as the uttermost goal of the research. In this research, a method based on texture feature for automatic identification of circular diatom images is presented. We aimed in this research to find the exact location of diatom by using the image segmentation and segmentation adjustment first, and then obtain eigenvectors by Fourier spectrum features, and finally use a BP neural network to effectively classify the diatoms with circular shape. With classification carried out using a BP neural network we attained 94.4% accuracy from a set of image containing twelve spieces of circular diatom. These species include mainly: Coscinodiscus oculatus (Fauv.) Petit, Coscinodiscus kuetzingii A. Schmidt, Coscinodiscus radiatus Ehrenberg, Coscinodiscus asteromphalus var. asteromphalus Ehrenberg, Coscinodiscus excentricus Ehrenberg, Coscinodiscus curvatulus var. curvatulus Grunow, Coscinodiscus wittianus Pantocsek, Arachnoidiscus Bailey, Cyclotella Kutzing, Actinoptyclrus Ehrenberg, Actinocyclus Ehrenberg, Hyalodiscus Ehrenberg. The result is an effective attempt of circular diatom identification based on texture character.