Plankton Portal (http://www.planktonportal.org/) is an online, citizen-science site that was developed in collaboration with Zooniverse to crowd-source classifications of plankton image collected by the DPI (formerly ISIIS). It also serves as an outreach and education platform, since scientists from our lab post regular social media and blog posts, and interact with participants on the extensive forum page. The original Plankton Portal site launched September 2013, using a small subset of (300,000) of images collected by our lab off the coast of California. Recently, we collaborated with French scientists from the University of Pierre and Marie Curie to release Plankton Portal 2.0 (launched June 10, 2015), with a revamped interface and an additional data set from Mediterranean Sea.
Since September 2013, over 8,400 unique citizen scientists have performed nearly 1 million classifications using Plankton Portal. Since each image is receives between 3-20 classifications from different users, this amounts to approximatley 405,000 images classified. In a recent month, the site has received over 55,000 page views. However, what is more telling of Plankton Portal's reach are the engagement statistics: there are over 28,000 by 500+ participants on the discussion forum, which include questions on the site, image classification, and plankton biology and ecology.
Cowen RK, Guigand C, Luo JY, Greer AT, Grassian B. Plankton Portal: an online science project for plankton classification and education. ASLO Ocean Sciences Meeting. 2014.
Luo JY, Robinson KL, Cowen RK, et al. A tale of two crowds: crowdsourcing for classification and machine learning algorithms in plankton imaging. In Prep.
In 2014-2015, we co-hosted the inaugural National Data Science Bowl (NDSB). The NDSB evolved from a collaboration between our lab at OSU, Booz Allen Hamilton (BAH), a management consulting firm, and Kaggle, a data science company. The aim of the NDSB was to challenge the international data science community to develop state-of-the art, predictive computer algorithms that would accurately classify DPI plankton images.
The competition ran from December 15, 2014 -March 16th, 2015. Over 1,000 participating teams collectively submitted more than 15,000 solutions to classify more than 100,000 images. The winning team, named “Deep Sea,” was a group of postdocs and graduate students from the University of Ghent in Belgium. The predictive algorithm they created had an average accuracy of 81% across all 121 plankton classes.
In return for providing the competition data set, our lab received Team Deep Sea’s winning solution as well as the other the top nine solutions. We are currently working on implementing these programs so we can use them to classify last year’s and this year’s DPI data. The data science community benefited as well. Tutorials and sample code were used extensively for learning and skills development and insights from the competition helped advance the state of the art in computer vision and Deep Learning. The archive competition data set is publically available for download and is being stored in perpetuity at NOAA’s National Oceanographic Data Center:
National Data Science Bowl Data Set Citation & Access: Cowen, Robert K.; Sponaugle, S.; Robinson, K.L.; Luo, J.; Oregon State University; Hatfield Marine Science Center (2015). PlanktonSet 1.0: Plankton imagery data collected from F.G. Walton Smith in Straits of Florida from 2014-06-03 to 2014-06-06 and used in the 2015 National Data Science Bowl (NODC Accession 0127422). NOAA National Centers for Environmental Information. Dataset. doi:10.7289/V5D21VJD
Please visit our lab's YouTube channel OSU Plankton Lab to check out cool videos.
Twitter: @OSUPlanktonLab. https://twitter.com/OSUPlanktonLab