Remote Sensing Transfer Learning for Species Distribution Modeling
Species distribution models link species observations to environmental variables. Current methods for collecting environmental variables generally consist of field-based methods or basic summary statistics of remotely sensed data. As the spatial extent of models expand to map continental and global patterns, field-based methods are no longer be feasible. While remotely sensed data offer a powerful alternative to physically collected features, summary statistics are rudimentary when compared to state-of-the-art computer vision techniques. We propose to replace coarse summaries with deep features extracted from high-resolution aerial images to better characterize habitats. We use a transfer learning approach to tune a Convolutional Neural Network to learn features relevant to species distribution modeling. We evaluate our approach using the citizen science based eBird dataset.
Major Advisor: Rebecca Hutchinson
Committee: Fuxin Li
Committee: Alan Fern
Monday, December 3, 2018 at 9:00am to 11:00am
Kelley Engineering Center, 1005
110 SW Park Terrace, Corvallis, OR 97331