Pollen Grains Detection, Segmentation, and Categorization
Shu Kong, Surangi W. Punyasena, Charless Fowlkes
latest updat: July 24, 2016 (dataset and code are released for the project of 3-way fossilized pollen!)
Detecting and classifying pollen grains in a collected sample allows one to estimate the diversity of plant species in a particular area. This is interesting for ecological monitoring (by placing pollen traps in different areas of the world) as well as in examining fossilized pollen to determine what plant species were present at a particular time in the past. We can easily collect and image many pollen samples but identifying and counting by eye the number of grains of each species is painstaking work which we would like to automate.
Specifically, we have the following projects
-
Identifying three species of fossil pollen grains from spruce.
-
Classifying large-scale modern pollen grains.
-
Identifying fossil pollen grains based on knowledge of modern pollen grains.
This project page is always being updated, please stay tuned!
Quick demonstration: focus adjustment, detection, segmentaion, classification, 3D modeling, matching
Dataset Release
-
Three-species fossilize spruce pollen dataset [here] (one species has not been released yet due to interest conflict)
-
large-scale modern pollen dataset [TBA]
-
modern and fossil spruce pollen dataset [TBA]
Code Release
-
demo of exemplar selection in the project of fossil pollen identification at species level [link]
-
demo of viewpoint aligment based on k-medoid clustering [link]
-
demo of effectiveness of patch-match method [link]
-
demo of how to select patches from pollen grain images [link]
Reference
-
Derek S. Haselhorst, Shu Kong, Charless C. Fowlkes, J. Enrique Moreno, David K. Tcheng, Surangi W. Punyasena, "Automating tropical pollen counts using convolutional neural nets: from image acquisition to identification", the iDigBio inaugural conference, 2017.
-
Surangi W. Punyasena, Shu Kong, Charless C. Fowlkes, and Stephen P. Jackson, "Reconstructing the extinction dynamics of Picea critchfieldii – the application of computer vision to fossil pollen analysis ", the iDigBio inaugural conference, 2017.