PALYIM: A DEEP-LEARNING PLATFORM FOR POLLEN IMAGE ANALYSIS

University Of Illinois At Urbana-Champaign, Urbana IL

Investigators

Abstract

Pollen and spores isolated from modern environmental and geologic samples are a critical data source for a broad range of fields in research and industry. Pollen is widely used to determine provenance and history in forensic analysis and archeological research. Fossil pollen is used to reconstruct ancient terrestrial environments and changes in paleoclimate. It is used to study the timing of plant originations and extinction and the evolution of plant biodiversity. Fossil pollen samples are also used to date the age of geologic sediments and are a critical resource for hydrocarbon exploration. However, despite the widespread use of pollen in biological and geological research, pollen identification remains a highly specialized, time-intensive, and primarily visual skill. This has impeded progress in the many fields that rely on these data. Automated the analysis of pollen samples, therefore, would vastly improve the quality and consistency of pollen data available for multiple areas of scientific research. This project will develop an intelligent web-accessible palynology image analysis platform (PALYIM) for hosting published computer vision tools for automated pollen identification. PALYIM will serve as a centralized repository of vetted, reproducible machine learning workflows specific to pollen identification and classification, drawing attention to the successes of this emerging field of research and crediting innovators within the community. Consolidating these efforts will allow us to build upon and expand the community knowledgebase. It will serve as a user-friendly gateway to machine learning analysis for the diverse communities that employ pollen data and automate and streamline curation of modern reference and fossil specimen images and image analysis workflows. The result will be a community platform that is accessible to researchers without experience in programming or machine learning. The project will increase the efficiency, replicability, and transparency of palynological research by allowing collaboration on a global scale. The platform will transform the dynamics of pollen research by incentivizing the sharing of analysis results and images by integrating public archival into the analysis workflow. Automation will reduce the research effort needed for pollen analyses and as a result will further encourage data sharing and large-scale analyses. The proposed project represents a unique in-demand and interdisciplinary training opportunity for early-career researchers and students who will participate in the development of the platform. This award reflects NSF's statutory mission and has been deemed worthy of support through evaluation using the Foundation's intellectual merit and broader impacts review criteria.