Micro-dumps detection in satellite images with Scattering Transform

Illegal landfills are unfortunately a plague in some areas of Southern Italy and cause significant direct and indirect environmental issues, including contamination of aquifers and dioxin release in the atmosphere due to arson attacks, potentially dangerous for local populations. The Italian government, at both central and local scale, is by long time enforcing these crimes, but the surveillance of wide areas is difficult with traditional on-site methodologies, even because it is possible that landfills are located within private properties not accessible without formal authorization. Therefore, remote sensing technologies are a key to improve and make more efficient the and frequent the monitoring activities.

Crowd for the Environment (C4E) is a project funded by the Italian Ministry of University and Research, aiming at the development of an innovative framework for the identification of illegal landfills using satellite and drone remote sensing in order to support decision makers in the organization of subsequent on-site actions. To this end, a new algorithm to detect possible polluted sites or sources of pollution has been developed. Specifically, the work has been focused on two particularly challenging and inter-related targets like micro-landfills and greenhouses.

Micro-landfills are often the result of waste disposal processes from industrial or agricultural activities partially or totally clandestine. The corresponding unregistered industrial or agricultural plants are potential sources of pollution. The comparison of satellite detections with the database of legal activities allowed to determine whether or not a plant is registered and therefore potentially harmful. In the study area, located in the nearby of the city of Caserta, greenhouses are a typical example of unregistered agricultural infrastructures which could illegally dispose micro-dumps of their plastic cover after the use.

Among the tested algorithms, those working on the spatial characterization of targets based on Scattering Transform were of particular interest. Such algorithms were used to extract textural features from images and their effectiveness was tested in comparison and in conjunction with spectral features within multi-class classifiers. The results obtained on very-high resolution Pleiades images with 50 cm spatial resolution showed that these features can significantly improve the detectors of both the identified targets.  In the case of greenhouses, which are targets without significant spectral characteristics, due to their transparency and reflectivity, the features based on Scattering Transform, alone, allow to build very competitive detectors. In the case of micro-dumps, which are targets very difficult to detect from satellite, both for their size and for the heterogeneous spectral characteristics, the use of the Scattering Transform seems the most effective tool, while the combined use of spectral features does not provide particular added value.

Ultimately, the use of the Scattering Transform seems to find an interesting application in the detection of environmental criticalities, also in relation to targets which are particularly difficult to be detected due to their high spectral ambiguity.