Bees provide essential pollination services to many crops and wild plants, but bees also need to take some of those plants' pollen in order to produce healthy offspring. The nutritional content of pollen varies dramatically from one species of flower to another. Unfortunately, since humans have changed the abundance and diversity of flowering resources in landscapes across the world, bees often struggle to find the nutrients that they need. Bee researchers often take pollen collected by bees and identify it under the microscope to better understand bees' dietary preferences and needs. However, the traditional approach to pollen ID takes a very long time, requiring trained experts to carefully compare the structures of bee-collected pollen grains with reference slides of pollen from known species. Machine learning can make the process of IDing bee-collected pollen vastly more efficient.
This year we plan to complete a pilot study using pollen collected from the California Botanic Garden's herbarium specimens, including each of the flower species commonly found at the Bernard Field Station (BFS). We aim to: 1) generate a large set of high-quality reference images, 2) train a model to automatically identify pollen from those species, 3) assess the model's accuracy with a subset of reference images, and 4) test it with samples collected from honey bee colonies at the BFS. These results will help us estimate how many training images we would need to identify pollen collected by bees across a more diverse region.
The plan is to have two students work on different aspects of the project. One student will optimize lab protocols and work on generating high-quality training/testing images. The other student will use computer vision techniques to process those images and train/assess a model to identify them as accurately as possible. Both students will meet together regularly to discuss progress and exchange ideas. Students in Prof Wloka's Spring 2022 Computer Vision course made great progress testing some approaches using a small subset of reference images, and we will be building off of that work. (If you are one of the students who did that work and you're interested in continuing it this fall, please apply!)
You will be part of an interdisciplinary group that includes students from Biology, CS, Math, Engineering, and other fields studying a wide variety of questions about collective decision-making in both bees and ants. In this lab you will develop new skills and get a sense of many different kinds of research and approaches to answering questions. You could also potentially continue your work in a senior thesis project, present at regional or national conferences, and/or co-author a publication.