Automated identification and classification of diatoms from water resources

Abstract

The quantity of certain types of diatoms is used for determining water quality. Currently, a precise identification of species present in a water sample is conducted by diatomists. However, different points of view of diatomists along with different sizes and shapes that diatoms may have in samples makes diatoms identification difficult, which is required to classify them into genera to which they belong to. Additionally, chemical processes, that are applied to eliminate unwanted elements in water samples (debris, flocs, etc.) are insufficient. Thus, diatoms have to be differentiated from those structures before classifying them into a genus. In fact, researchers have a special interest on looking for different ways to perform an automated identification and classification of diatoms. In spite of applications, an automatic identification of diatom has a high level of difficulty, due to the present of unwanted elements in water samples. After diatoms have been identified, diatoms classification into genera is an additional problem. In this paper, an automatic method for identification and classification diatoms from images is presented. The method is based on the combination of Scale and Curvature Invariant Ridge Detector (SCIRD-TS), following by a post processing method, and the use of a nested Convolutional Neural Networks (CNN). Whilst the identification approach is able to identify well-defined ridge structures, the nested CNN is able to classify a diatom into the genus to which it belongs to.