Automated Fish Classification | SRI International

Toggle Menu

Automated Fish Classification

FLASK provides a fast annotation and machine learning based classification capability.

For the NOAA Fisheries Stock Assessment Strategic Initiative program, SRI researchers developed FLASK, a semi-automated annotation tool that works with minimal user involvement.  Using FLASK, analysts can rapidly annotate video that is then used to iteratively train a neural network to correctly classify a moving image.

FLASK’s neural network framework preprocessing functionality extracts key object features, segments images and clusters fish types and other undersea objects, enabling the rapid annotation and training engine within FLASK to continually learn information.  After training, the FLASK-created neural network is used for performing rapid classification of raw video sources with specific fish types and counts.

The capabilities enabled by FLASH for fish classification are extensible to a wide range of applications and needs.  FLASK is applicable to many applications such as surveillance (people, vehicles, objects), and inspection (agricultural, assembly line) applications.

This effort was sponsored under Award Number NA14OAR4320260 by NOAA and through Florida Atlantic University.