Serial image analysis to infer spatially-explicit species dynamics

Develop, train, and improve upon machine learning algorithms to identify and enumerate species in a large collection of fixed-location photographs that were taken repeatedly (on a monthly basis) over three years in an experiment of ecological succession in the Oregon marine intertidal.

Objectives


The images whose species require enumeration represent 108 fixed-location time-series (each ~36 months long for a total of ~3,888 images) in which the identity of ten to several hundreds of individuals requires determination. With some individuals surviving (and growing) over the entire time-series and others being present in only a single photograph, the time-series nature of these images offers both informative information on which to capitalize and challenges associated with the probabilistic prediction of rare occurrences.  Key to success will be the optimized utilization of a painstakingly amassed set of 1,944 "training set" images in which the identity and precise x-y location of each individual has been recorded "by hand".  Building on the successes of prior EECS Capstone teams, progress towards achieving the goals of this project will lead the way for powerful statistical approaches with which to understand the dynamics of complex ecological systems, for which the current “training” dataset alone is poorly suited.

Motivations


The overarching motivation of the here-proposed project is to help disentangle the role of species interactions in driving the abundance dynamics of a species-rich community in the Oregon marine intertidal. Understanding how and why species abundances vary in space and over time represents a fundamental and pressing challenge for the field of Ecology.  Many processes influence species abundances, including both extrinsic disturbances and processes such as storm events, seasonality, and climate change, and intrinsic processes such as birth-death events, predator-prey interactions, and competition for limited resources. These multi-dimensional aspects inherent to species-rich ecological communities mirror the processes that underlie complex adaptive systems in general.  

Qualifications


Minimum Qualifications:
None Listed

Preferred Qualifications:

Students who have taken some or all of the core AI courses will be best-equipped for this project. 


CS 331 Introduction to Artificial Intelligence (4)
CS 434 Machine Learning and Data Mining (4)
MTH 254 Vector Calculus I (4)
MTH 341 Linear Algebra I (3)
ST 421 Introduction to Mathematical Statistics I (4)
CS 475 Introduction to Parallel Programming (4)


Details


Project Partner:

Mark Novak

NDA/IPA:

No Agreement Required

Number Groups:

1

Project Status:

Accepting Applicants

Website:
https://github.com/NovakLab-EECS/OR_Intertidal_ExpPatch_ImageAnalysis
Card Image Capstone