A Visual System for Autonomous Foraminifera Identification
The goal of this project is to develop an automated system for identification of foraminifera (single-celled organisms with shells). Currently undergraduate workers are often employed to hand pick several thousands of specimens from ocean sediments for each study. This is tedious and time consuming work. By automating the bulk of the identification process, user expertise can be focused on verification and identification of subtle differences.
A visual identification system will be developed in order to automate the identification of target microorganisms. The visual system will incorporate a controllable LED lighting ring used to capture images by illuminating the specimens from several directions, mimicking an important step in the traditional identification process. These images will be used to create a 3D model of the organism in real-time within a second. Computer vision and pattern recognition techniques will be tuned to acceptable recognition rates set by feedback from an expert in paleoceanography who will also provide labeled samples for training and validation. The initial proof of concept study will focus on identifying six species of planktonic foraminifera, and their morphotypes, that are widely used by paleoceanographers.
This work is a collaboration between the ARoS Lab, Dr. Thomas M. Marchitto (Associate Professor at the University of Colorado Boulder) and Dr. Ritayan Mitra (Assistant Professor at IIT Bombay). This work is funded by the NSF under award OCE-1637039 (August 2016 – July 2019).
In the News
- February 6, 2019 – ScienceDaily – Artificial Intelligence can Identify Microscopic Marine Organisms
- July 29, 2016 – Inverse – What Microscopic Shell Fossils Can Teach Us About Ancient Oceans
- July 5, 2016 – NCSU News – How Teaching Robots to Identify Microscopic Fossils Could Help Us Understand Oceans
- NCSU-AROS Foraminifera Identification GitHub [Released: March 2019] – A Keras implementation for foraminifera identification. For questions about this code please contact Dr. Edgar Lobaton. Please cite  if using this model.
- NCSU-CUB Foram Images 01 [Released: October 2017] – Contains images of over 1,000 forams taken under 16 different lighting directions. The species and locations of the samples are also specified. Please cite  if using this dataset in your research.
- NCSU-CUB Foram Labels 01 [Released: October 2017] – Contains manual segmentation of over 400 samples from the NCSU-CUB Foram Images 01 dataset. The segmentation labels are matched by their name. Please cite  if using this dataset in your research.
- R. Mitra, T.M. Marchitto, Q. Ge, B. Zhong, B. Kanakiya, M.S. Cook, J.S. Fehrenbacher, J.D. Ortiz, A. Tripati, E. Lobaton, “Automated species-level identification of planktic foraminifera using convolutional neural networks, with comparison to human performance,” Marine Micropaleontology, 147, 2019, pp 16-24.
- B. Zhong, Q. Ge, B. Kanakiya, R. Mitra, T. Marchitto, E. Lobaton, “A Comparative Study of Image Classification Algorithms for Foraminifera Identification,” IEEE Symp. Series on Computational Intelligence (SSCI), 2017. [Makes use of NCSU-CUB Foram Images 01 Dataset]
- Q. Ge, B. Zhong, B. Kanakiya, R. Mitra, T. Marchitto, E. Lobaton, “Coarse-to-Fine Foraminifera Image Segmentation through 3D and Deep Features,” IEEE Symp. Series on Computational Intelligence (SSCI), 2017. [Makes use of NCSU-CUB Foram Labels 01 Dataset]