Automated Plankton Image Classification

The overall goal of this project was to process a digital image data collected using the In Situ Ichthyoplankton Imaging System (ISIIS) during the summer of 2011. These data were collected on three separate cruises and total roughly 50 terabytes (TB) of image data, corresponding to approximately 10,700,500 individual images requiring analysis. We have developed an approach (and system) that successfully identified and quantified approximately 160,000,000 segmented images (contained within the 10,700,500 larger images), or Regions of Interest (ROIs). These ROI are either individual plankton species or other material (such as detritus). Examples of the extracted and quantified images can be seen below.

Technology

MongoDB, C++, Matlab, Python, PostGIS