Objective comparison of relief visualization techniques with deep CNN for archaeology

Archaeology has been profoundly transformed by the advent of airborne laser scanning (ALS) technology (a.k.a LiDAR). High-resolution and high-precision synoptic views earth’s topography are now available, even in densely forested environments, to identify characterize landform patterns resulting from past human occupation. ALS-based archaeological prospection relies on digital terrain model (DTM) visualization techniques (VTs) that highlight subtle topographical changes perceived interpreted archaeologists. An increasing number VTs have developed, they evaluated date mainly based subjective perception. This study developed a new approach state-of-the-art computer-vision algorithms benchmark using objective metrics. Thirteen were applied ALS-derived DTM, deep convolution neural network (deep CNN) was implemented trained automatically detect segment structures these images. Visual interpretation images showed most informative VT e2MSTP, which combined multiscale topographic analysis (MSTP) with morphologically explicit image slope-invariant relief detrending technique. The CNN confirmed results provided performance indicates computer vision opens perspectives selection suitable for prospection.