Presentation Title

Using machine learning to classify and estimate relative abundances of heterotrophic protists

Faculty Mentor

Darcy Taniguchi

Start Date

23-11-2019 10:45 AM

End Date

23-11-2019 11:30 AM

Location

116

Session

poster 4

Type of Presentation

Poster

Subject Area

biological_agricultural_sciences

Abstract

Heterotrophic protists serve a vital ecological role within marine food webs by acting as the major consumers of primary producers and serving as an essential food source for higher trophic levels. Despite their ecological importance, being able to measure the abundances of heterotrophic protists has proven to be an arduous task due to their small sizes (~10-100 um) and lack of pigments. Currently, techniques to measure the abundances of heterotrophic protists rely on sampling by bottles and microscope analysis, an approach that’s time consuming and may potentially damage live specimen. To automate the process of measuring the relative abundances of heterotrophic protists without damaging them and to study their relationship with autotrophic prey, we utilize a machine learning tool called Convolutional Neural Networks (CNNs). CNNs are trainable machine learning algorithms that can be used for image classification. A microscope system located at the Scripps Memorial Pier, called the Scripps Plankton Camera System (SPCS), performs real-time in situ imaging of drifting organisms found within the waters of La Jolla, California. Using the images from this system, we have created a training set of labeled heterotrophic protists, enabling the means for us to train a CNN that can accurately classify images taken from SPCS into predetermined categories. Using the network with the highest accuracy, we can input images from the SPCS to classify organisms and ultimately estimate the relative abundances of heterotrophic protists as they change overtime. This information can then be used to address important ecological questions such as the role of heterotrophic protists in the formation and decline of algal blooms and simultaneously help advance the field of machine learning

This document is currently not available here.

Share

COinS
 
Nov 23rd, 10:45 AM Nov 23rd, 11:30 AM

Using machine learning to classify and estimate relative abundances of heterotrophic protists

116

Heterotrophic protists serve a vital ecological role within marine food webs by acting as the major consumers of primary producers and serving as an essential food source for higher trophic levels. Despite their ecological importance, being able to measure the abundances of heterotrophic protists has proven to be an arduous task due to their small sizes (~10-100 um) and lack of pigments. Currently, techniques to measure the abundances of heterotrophic protists rely on sampling by bottles and microscope analysis, an approach that’s time consuming and may potentially damage live specimen. To automate the process of measuring the relative abundances of heterotrophic protists without damaging them and to study their relationship with autotrophic prey, we utilize a machine learning tool called Convolutional Neural Networks (CNNs). CNNs are trainable machine learning algorithms that can be used for image classification. A microscope system located at the Scripps Memorial Pier, called the Scripps Plankton Camera System (SPCS), performs real-time in situ imaging of drifting organisms found within the waters of La Jolla, California. Using the images from this system, we have created a training set of labeled heterotrophic protists, enabling the means for us to train a CNN that can accurately classify images taken from SPCS into predetermined categories. Using the network with the highest accuracy, we can input images from the SPCS to classify organisms and ultimately estimate the relative abundances of heterotrophic protists as they change overtime. This information can then be used to address important ecological questions such as the role of heterotrophic protists in the formation and decline of algal blooms and simultaneously help advance the field of machine learning