USING MACHINE LEARNING TO DISCERN ANTHROPOGENIC FEATURES WITHIN LIDAR DATA SETS

The detailed topographic data provided by Lidar gives us the opportunity to identify archaeological features in landscapes where access is difficult either due to terrain, vegetation, or site dimensions. These data have been used to successfully locate new archaeological sites and give a new perspective on known sites. Typically this work is done manually, where topographic data is searched to find areas of interest, yet this can be time intensive and subjective. Our interdisciplinary team of geologists, archaeologists, and computer scientists are working on developing a technique to automatically extract features of interest by utilizing a training data set of known features. For this project we are using Lidar data from American Samoa and North Dakota because these two locations are studied by our team and provide significant contrast to ensure that the technique works in a variety of locations. Training data sets have been developed for both locations by carefully tracing known features. A moving window scaled to the feature of interest is then passed through the training data and used to identify the characteristics of the known features. The same sized window is later passed over areas where features are likely but have not yet been found to identify previously unknown features. Preliminary results suggest that this approach can successfully be used to identify archaeological features in a variety of landscapes.