Classification of Radiolarian Fossil Images with Deep Learning Methods


KEÇELİ A. S. , UZUNÇİMEN KEÇELİ S., KAYA A.

26th IEEE Signal Processing and Communications Applications Conference (SIU), İzmir, Turkey, 2 - 05 May 2018 identifier identifier

  • Publication Type: Conference Paper / Full Text
  • Volume:
  • Doi Number: 10.1109/siu.2018.8404460
  • City: İzmir
  • Country: Turkey

Abstract

Radiolarians are some kind of planktonic protozoa and are important biostratigraphic and paleoenvironmental indicators for palaeogeographic reconstructions. Radiolarian paleontology is still considered to be the most affordable way to date deep ocean sediments. Conventional methods for identifying radiolarians are time consuming and can not be scaled by the detail or scope required for large-scale studies. Automatic image classification allows these analyzes to be done quickly. In this study, a method for automatic classification of fosilized radiolarian images obtained by Scanning Electron Microscope (SEM) has been proposed. The study included a Convolutional Neural Network (CNN) trained using radiolarian images and another CNN model that was pre-trained and fine-tuned. High classification performances were obtained with the generated models. It has been observed that the results obtained from fine-tuned model are more successful.