Detection of Terrain Structures in Airborne Laser Scanning Data Using Deep Learning

Abstract

Abstract. Automated recognition of terrain structures is a major research problem in many application areas. These structures can be investigated in raster products such as Digital Elevation Models (DEMs) generated from Airborne Laser Scanning (ALS) data. Following the success of deep learning and computer vision techniques on color images, researchers have focused on the application of such techniques in their respective fields. One example is detection of structures in DEM data. DEM data can be used to train deep learning models, but recently, Du et al. (2019) proposed a multi-modal deep learning approach (hereafter referred to as MM) proving that combination of geomorphological information help improve the performance of deep learning models. They reported that combining DEM, slope, and RGB-shaded relief gives the best result among other combinations consisting of curvature, flow accumulation, topographic wetness index, and grey-shaded relief. In this work, we approve and build on top of this approach. First, we use MM and show that combinations of other information such as sky view factors, (simple) local relief models, openness, and local dominance improve model performance even further. Secondly, based on the recently proposed HR-Net (Sun et al., 2019), we build a tinier, Multi-Modal High Resolution network called MM-HR, that outperforms MM. MM-HR learns with fewer parameters (4 millions), and gives an accuracy of 84:2 percent on ZISM50m data compared to 79:2 percent accuracy by MM which learns with more parameters (11 millions). On the dataset of archaeological mining structures from Harz, the top accuracy by MM-HR is 91:7 percent compared to 90:2 by MM.

Automated large‐scale mapping and analysis of relict charcoal hearths in Connecticut (USA) using a Deep Learning YOLOv4 framework

“…We are aware that multiscale visualization techniques can result in better detection results when using Deep Learning (Guyot et al, 2021). Even though, SLRM produces relatively high results in a recent study comparing visualizations (Guyot et al, 2021) as well as in less recent publications (Gallwey et al, 2019;Kazimi et al, 2020). More importantly, we choose efficiency over total accuracy as we are dealing with (very) large-scale datasets.…”

A modified Mask region‐based convolutional neural network approach for the automated detection of archaeological sites on high‐resolution light detection and ranging‐derived digital elevation models in the North German Lowland

“…Since high-resolution LiDAR data have become readily available for an increasing number of countries and RCH sites are found in increasing larger forested areas, fully or semiautomated methods are required to decrease the workload of such labour-intensive manual mappings. Some previous (semi)automated mapping approaches for RCHs and other archaeological objects involve template matching (Schneider et al, 2014), geographic object-based image analysis (GEOBIA) (Witharana et al, 2018) and, more recently, deep learning techniques (e.g., Kazimi et al, 2020;Lambers et al, 2019;Trier et al, 2018;Trier et al, 2021;Verschoof-van der Vaart et al, 2020).…”

Automated methods for image detection of cultural heritage: Overviews and perspectives

“…These authors took advantage of the fact that some anthropogenic or natural buried structures such as ditches and walls create spatial and spectral anomalies which can be perceived from the way in which the objects and their background appear in the image. Among the different approaches for automated classification, ML‐based experiments are increasingly on the rise and there is clearly a great emphasis on methodologies applying ML that have appeared in the last 15 years (Lasaponara et al, 2014; Lasaponara et al, 2016; Menze et al, 2006), and more recently on using deep learning architectures (Verschoof‐van der Vaart & Landauer, 2021; Guyot et al, 2018; Kazimi et al, 2018; Lambers et al, 2019; Trier et al, 2016; Zingman et al, 2016). In fact, for certain objects, and given sufficient data, DL approaches can achieve or exceed 80% accuracy (Lambers et al, 2019; Trier et al, 2016).…”

“…Regardless of the approach used to work with the data, the interpretation and validation of objects extracted from images is the archaeologist's prerogative (Quintus et al, 2017; Traviglia et al, 2016; Verdonck et al, 2017). In any case, in order to set up automated earth image analysis/detection, it is necessary to start by feeding the automated systems with a large amount of data, which would be difficult to visualize manually in the first place (Kazimi et al, 2018), given that over the last century a gigantic ‘black box’ of data has been collected and stored but is yet to be visualized.…”