Semantic Segmentation of Airborne LiDAR Data in Maya Archaeology

Marek Bundzel,Tomáš Tkáčik,Tibor Lieskovský,Miroslav Jaščur,Milan Kováč,Peter Sinčák

doi:10.3390/rs12223685

Abstract

Airborne LiDAR produced large amounts of data for archaeological research over the past decade. Labeling this type of archaeological data is a tedious process. We used a data set from Pacunam LiDAR Initiative survey of lowland Maya region in Guatemala. The data set contains ancient Maya structures that were manually labeled, and ground verified to a large extent. We have built and compared two deep learning-based models, U-Net and Mask R-CNN, for semantic segmentation. The segmentation models were used in two tasks: identification of areas of ancient construction activity, and identification of the remnants of ancient Maya buildings. The U-Net based model performed better in both tasks and was capable of correctly identifying 60–66% of all objects, and 74–81% of medium sized objects. The quality of the resulting prediction was evaluated using a variety of quantifiers. Furthermore, we discuss the problems of re-purposing the archaeological style labeling for production of valid machine learning training sets. Ultimately, we outline the value of these models for archaeological research and present the road map to produce a useful decision support system for recognition of ancient objects in LiDAR data.

Highlights

In 2016, the Pacunam LiDAR Initiative (PLI) undertook the largest LiDAR survey to date of the lowland Maya region, mapping a total of 2144 km2 of the Maya Biosphere Reserve in Petén, Guatemala [1]
We consider the U-Net model that uses 30 × 30 averaging box filter applied to OT and IoU optimizing threshold for binary segmentation to be the best performing model, outperforming Mask R-Convolutional Neural Networks (CNNs) model by a narrow margin
Mask region-based CNNs (R-CNN) was trained for 100 epochs, no output smoothing was used

Summary

Introduction

In 2016, the Pacunam LiDAR Initiative (PLI) undertook the largest LiDAR survey to date of the lowland Maya region, mapping a total of 2144 km of the Maya Biosphere Reserve in Petén, Guatemala [1] The result of this endeavor is a standardized and consistent data set that enables archaeological studies of an area covered by tropical forest that limits the on-the-ground research. Based on the experimental results, we propose how to build a decision support system that aids object labeling

Objectives

Methods

Findings

Discussion

Conclusion