Gianni Cristian Iannelli

Building footprint detection and outlining from satellite imagery represents a very useful tool in many types of applications, ranging from population mapping to the monitoring of illegal development, from urban expansion monitoring to organizing prompter and more effective rescuer response in the case of catastrophic events. The problem of detecting building footprints in optical, multispectral satellite data is not easy to solve in a general way due to the extreme variability of material, shape, spatial, and spectral patterns that may come with disparate environmental conditions and construction practices rooted in different places across the globe. This difficult problem has been tackled in many different ways since multispectral satellite data at a sufficient spatial resolution started making its appearance on the public scene at the turn of the century. Whereas a typical approach, until recently, hinged on various combinations of spectral–spatial analysis and image processing techniques, in more recent times, the role of machine learning has undergone a progressive expansion. This is also testified by the appearance of online challenges like SpaceNet, which invite scholars to submit their own artificial intelligence (AI)-based, tailored solutions for building footprint detection in satellite data, and automatically compare and rank by accuracy the proposed maps. In this framework, after reviewing the state-of-the-art on this subject, we came to the conclusion that some improvement could be contributed to the so-called U-Net architecture, which has shown to be promising in this respect. In this work, we focused on the architecture of the U-Net to develop a suitable version for this task, capable of competing with the accuracy levels of past SpaceNet competition winners using only one model and one type of data. This achievement could pave the way for achieving better performances than the current state-of-the-art. All these results, indeed, have yet to be augmented through the integration of techniques that in the past have demonstrated a capability of improving the detection accuracy of U-net-based footprint detectors. The most notable cases are represented by an ensemble of different U-Net architectures, the integration of distance transform to improve boundary detection accuracy, and the incorporation of ancillary geospatial data on buildings. Our future work will incorporate those enhancements.

There is an increasing trend at using social media data to map human activity, to the point that some authors have suggested that these data may be even better than Earth Observation (EO) data to map urban areas. This work introduces a novel approach to map urban areas using SAR images from Sentinel-1 data, and exploiting either Twitter or Weibo data for two different cities, Beijing and Taipei. The use of different social media data results into different urban extent maps, and for different cities the same approach provides different accuracy values for the same area, according to the availability of either social media data set. Instead, urban extents obtained by means of EO data only are more stable. Indeed, the adoption of a specific social platform depends on a number of geographical, economic and political factors, while remote sensing is based on physics.

This paper illustrates a fusion approach to jointly exploit Sentinel-1 (S1) and Sentinel-2 (S2) data to detect urban areas. The proposed procedure is designed to automatically and effectively exploit the specific characteristics of synthetic aperture radar (SAR) and multispectral data, so that it can be safely applied to different urban environments with satisfying results. To this aim, it starts from two previously developed algorithms for urban extent extraction designed for multispectral or SAR spaceborne data. They are first adapted to use S1 and S2 datasets, and then coupled to a novel procedure that merges the two processing chains with the goal to exploit the spectral properties of man-made materials as well as the double bounce backscatter effect that is common in built-up environments. Experimental results in urban areas all around the world show that the proposed approach is successful at combining S1 and S2 information by reducing the errors of the two original techniques.

Cash crops are agricultural crops intended to be sold for profit as opposed to subsistence crops, meant to support the producer, or to support livestock. Since cash crops are intended for future sale, they translate into large financial value when considered on a wide geographical scale, so their production directly involves financial risk. At a national level, extreme weather events including destructive rain or hail, as well as drought, can have a significant impact on the overall economic balance. It is thus important to map such crops in order to set up insurance and mitigation strategies. Using locally generated data—such as municipality-level records of crop seeding—for mapping purposes implies facing a series of issues like data availability, quality, homogeneity, etc. We thus opted for a different approach relying on global datasets. Global datasets ensure homogeneity and availability of data, although sometimes at the expense of precision and accuracy. A typical global approach makes use of spaceborne remote sensing, for which different land cover classification strategies are available in literature at different levels of cost and accuracy. We selected the optimal strategy in the perspective of a global processing chain. Thanks to a specifically developed strategy for fusing unsupervised classification results with environmental constraints and other geospatial inputs including ground-based data, we managed to obtain good classification results despite the constraints placed. The overall production process was composed using “good-enough" algorithms at each step, ensuring that the precision, accuracy, and data-hunger of each algorithm was commensurate to the precision, accuracy, and amount of data available. This paper describes the tailored strategy developed on the occasion as a cooperation among different groups with diverse backgrounds, a strategy which is believed to be profitably reusable in other, similar contexts. The paper presents the problem, the constraints and the adopted solutions; it then summarizes the main findings including that efforts and costs can be saved on the side of Earth Observation data processing when additional ground-based data are available to support the mapping task.

Cash crops are agricultural crops intended to be sold for profit as opposed to subsistence crops, meant to support the producer, or to support livestock. Since cash crops are intended for future sale, they translate into large financial value when considered on a wide geographical scale, so their production directly involves financial risk. At a national level, extreme weather events including destructive rain or hail, as well as drought, can have a significant impact on the overall economic balance. It is thus important to map such crops in order to set up insurance and mitigation strategies. Using locally generated data -such as municipality-level records of crop seeding- for mapping purposes implies facing a series of issues like data availability, quality, homogeneity etc. We thus opted for a different approach relying on global datasets. Global datasets ensure homogeneity and availability of data, although sometimes at the expense of precision and accuracy. A typical global approach makes use of spaceborne remote sensing, for which different land cover classification strategies are available in literature at different levels of cost and accuracy. We selected the optimal strategy in the perspective of a global processing chain. Thanks to a specifically developed strategy for fusing unsupervised classification results with environmental constraints and other geospatial inputs including ground-based data, we managed to obtain good classification results despite the constraints placed. The overall production process was composed using ``good-enough" algorithms at each step, ensuring that the precision, accuracy, and data-hunger of each algorithm was commensurate to the precision, accuracy, and amount of data available. This paper describes the tailored strategy developed on the occasion as a cooperation among different groups with diverse backgrounds, a strategy which is believed to be profitably reusable in other, similar contexts. The paper presents the problem, the constraints and the adopted solutions; it then summarizes the main findings including that efforts and costs can be saved on the side of Earth Observation data processing when additional ground-based data are available to support the mapping task.

In order for a risk assessment to deliver sensible results, exposure in the concerned area must be known or at least estimated in a reliable manner. Exposure estimation, though, may be tricky, especially in urban areas, where large-scale surveying is generally expensive and impractical; yet, it is in urban areas that most assets are at stake when a disaster strikes. Authoritative sources such as cadastral data and business records may not be readily accessible to private stakeholders such as insurance companies; airborne and especially satellite-based Earth-Observation data obviously cannot retrieve all relevant pieces of information. Recently, a growing interest is recorded in the exploitation of street-level pictures, procured either through crowdsourcing or through specialized services like Google Street View. Pictures of building facades convey a great amount of information, but their interpretation is complex. Recently, however, smarter image analysis methods based on deep learning started appearing in literature, made possible by the increasing availability of computational power. In this paper, we leverage such methods to design a system for large-scale, systematic scanning of street-level pictures intended to map floor numbers in urban buildings. Although quite simple, this piece of information is a relevant exposure proxy in risk assessment. In the proposed system, a series of georeferenced images are automatically retrieved from the repository where they sit. A tailored deep learning net is first trained on sample images tagged through visual interpretation, and then systematically applied to the entire retrieved dataset. A specific algorithm allows attaching “number of floors” tags to the correct building in a dedicated GIS (Geographic Information System) layer, which is finally output by the system as an “exposure proxy” layer.