All events are in Central time unless specified.
Seminar

PhD Dissertation Defense - Qiao Hu

Drone & Al for precision conservation: a case study in playa wetlands of the Rainwater Basin in Nebraska

Date:
Time:
12:00 pm – 1:00 pm
Hardin Hall Room: 207 South
3310 Holdrege St
Lincoln NE 68583
Additional Info: HARH
Contact:
Wayne Woldt, wwoldt1@unl.edu
Artificial Intelligence (AI) in computer vision is revolutionizing the geoscience and remote
sensing domains. Convolutional neural network (CNN), as a typical Automated machine
learning (AutoML), automatically represents image contexts (spatial-spectral correlation) from remote sensing imagery into spatially and spectrally relevant knowledge. The end-to-end design relieves researchers from tedious feature engineering and local tuning work.
Considering the vital role of long-term frequent land cover monitoring in assessing the local
ecosystem health cost-effective adaptations of AutoML, such as CNN, in local cases are vital to support in-time conservation and proactive ecological management. This study aims to explore the adaptation of these cutting-edge AutoML techniques in wetland monitoring and
delineation, which can improve traditional wetland mapping pipelines by facilitating cost
savings. Three publicly managed playa wetlands in the Rainwater Basin, Nebraska, USA, were selected as the study areas. By implementing AutoML techniques, I want to address three critical aspects in wetland ecosystem monitoring: 1) automatic waterfowl censusing using thermal sensors, 2) automatic wetland inundation delineation during the spring season using multi-spectral sensors, 3) and automatic aquatic vegetation segmentation under dynamic environments (during the fall season) using optical sensors. The remote sensing data in this study mainly relies on high spatial resolution UAV imagery. Different levels of computer vision techniques, including image processing, machine learning, and deep learning approaches, were developed and tested. The results indicate that CNN with proper designs and configurations can facilitate significant cost-savings on wetland mapping.

https://unl.zoom.us/j/98700954080

Download this event to my calendar

This event originated in School of Natural Resources.