Authors: Eli Egan-Anderson*, University of Connecticut, William B. Ouimet, University of Connecticut, Chandi Witharana, University of Connecticut
Topics: Remote Sensing, Geographic Information Science and Systems, Geomorphology
Keywords: Deep Learning, Remote Sensing, Land Use
Session Type: Poster
Start / End Time: 8:00 AM / 9:40 AM
Room: Lincoln 2, Marriott, Exhibition Level
Presentation File: No File Uploaded
Advanced machine learning combined with public, widely available, LiDAR data has great potential to automate the extraction and classification of historic land use features preserved under forest canopy throughout the northeast US landscape. Previous studies have shown that stone walls, house foundations and relict charcoal hearths (RCHs) stand out clearly in derivative LiDAR DEM products such as slope and hillshade maps but to date, mapping has been mainly carried out by on screen digitization. In this study, a deep learning convolutional neural network (CNN) algorithm was employed to extract relict charcoal hearth and stone wall features from LiDAR data. By employing the CNN algorithms in eCognition on filtered LiDAR slope maps (high pass and low pass) we were able to highlight distinctive aspects of these features to identify where they occur. The CNN model was trained using hand annotated datasets derived from 1m LiDAR DEMs. The model inferences were further refined using object-based image analysis methods. This approach offers value in speed, accuracy and ease as well as a unique insight into the extent and spatial distribution the past land use activities in Connecticut, as well as anywhere in the northeastern US where LiDAR data exists and the landscape experienced similar land use activity during the 17th-20th centuries.