Remote and Close Range Sensing for the Automatic Identification and Characterization of Archaeological Looting. The Case of Peru

Looting is the major source of artefacts for the antiquities market. Specific measures are needed to fight the whole chain of the illicit activities undertaken by criminal organizations (from the excavation to the selling of the artefacts), and they should be devised for each phase of such illegal activities. The development and use of appropriate technologies for the identification of the most 'vulnerable' sites, and the timely detection and automatic quantification of the extension of the looted areas are crucial steps for setting up a monitoring system working also for remote and inaccessible archaeological areas, often in regions affected by armed conflicts or characterized by flight restrictions. In this context, Earth Observation (EO) technologies can provide reliable information: (i) to quantify the looting phenomenon even if it is on an 'industrial scale' over large areas, and (ii) to set up a systematic monitoring tool to trace the illicit trade in antiquities. In this paper, an improvement of the Archaeological Looting Feature Extraction Approach (ALFEA) -developed by the same authors in 2018- is proposed to further improve the ability in the automatic identification and extraction of looting features for heterogeneous desert landscapes, characterized not only by looting patterns but also by archaeological micro-relief and emerging remains, as well as by natural geomorphological features and the presence of structures and dirt pathways, which exhibit a similar spectral behavior but dimensions, morphology, and/or geometric patterns different from those linked to looting. The improvement of ALFEA (ALFEA-I) was applied in significant test areas considered among the most important archaeological sites in Peru, (i) Pachacamac close to Lima, and (ii) Ventarron in the Lambayeque region Northern Peru. The first site is characterized by past clandestine excavations and looting is difficult to recognize both in situ and from satellite image; the second site is affected by more recent archaeological disturbances due to grave robberies, easier to identify from remote sensing data. The original ALFEA -composed of the sequential integration of spatial autocorrelation statistics, unsupervised classification, and segmentation- has been herein refined by adding a processing step based on multi-threshold parameters of segmentation, thus improving the performance in terms of extraction capability of looting features in case of heterogeneous areas. Tthe integration