A Study on Hierarchical Classification Model for Diatom Images using Deep Convolutional Neural Networks

Abstract

Diatoms are a huge and ecologically significant group of unicellular or colonial newlineorganisms (algae). Diatoms are, jointly with invertebrates are commonly employed newlineorganisms in the examination of river quality. Several works also support the newlineefficiency of biological indices depending upon diatoms for controlling the newlineecological status of water in rivers. The application of the most widespread diatom newlineindices always needs an accurate level of classification, which necessitates more time newlineand expert training. Besides, the recognition of morphological microstructures and newlinefrustules discrimination from other elements in the image is yet to be solved. Few newlinestudies have been available in the literature for the automatic classification of diatom newlineimages. But they are valid only for a restricted subset of species. Therefore, the newlinenumber of examined species has been restricted and the results are comparatively newlinelow, reducing the results with an increase in number of species. Considering the newlinelimitations existing in the state of art methods, the proposed research work the aims newlineto assist a taxonomist in identifying a wide range of different diatoms by developing newlinean automated hierarchical classification model for the effective classification of newlinediatom species and genus images. The entire research work is organized into a set of newlineresearch objectives as listed below newlineTo develop an Improved Canny Edge Detection Model based on edge newlinedetection technique to effectively detect the edges of the diatom images. To present an AlexNet Transfer Learning with Random Forest Classifier newline(ATLRFC) model as feature extractor and classifier for extracting applicable newlinefeatures and classify the diatom images effectively. To present an AlexNet Transfer Learning with Decision Tree Classifier newline(ATLDTC) model as a feature extractor and classifier in order to obtain the newlineapplicable features and classify diatom images efficiently. newline To design an Inception V4 Transfer Learning with Random Forest Classifier newline(ITLRFC) model for extracting the helpful