Identifying Iowa BMPs via Deep Learning Image Segmentation

Tuesday, January 28, 2020 - 1:00pm to 1:30pm
Event Type: 

Speaker: Charlie Labuzzetta

Abstract: A variety of Best Management Practices (BMPs) for water pollution control are installed on agricultural land throughout Iowa. Recently, the ISU GIS Facility finished the Iowa BMP Mapping Project, a GIS catalogue of BMPs installed on Iowa lands between the years of 2007 - 2010. This research focuses on developing a deep learning approach to image segmentation which can help to identify BMPs via remote sensing imagery in years outside of the database's range. This may help track the trends in the installation of these BMPs over time and offer a reduced cost method for updating the dataset. Last fall, I presented the theoretical framework of this deep learning approach and now I will present a variety of experimental results from this project. This talk will focus on the updated methodology for identifying grassed waterway features, but I will also present the preliminary results for other BMPs including terraces, pond dams, and WASCOBs