Overview and details of the sessions of this conference. Please select a date or location to show only sessions at that day or location. Please select a single session for detailed view (with abstracts and downloads if available).
Christian Doppler Laboratory GEOHUM, University of Salzburg, Austria
Satellite imagery is an important information source for research on remote sensing (RS)-based humanitarian applications. The selection of satellite imagery is one of the most important steps for such research. This paper firstly shows the selection of satellite imagery in past research from 2015 to 2021. It can be found that most research on land cover and land use (LCLU) change caused by conflicts or refugees/internally displaced persons (IDPs) chose medium spatial resolution (MSR) imagery. Most research on dwelling detection of refugee/IDP camps applied high or very high spatial resolution (HSR/VHSR) imagery. There is much research that applied multiple types of satellite imagery for humanitarian applications. Then, the paper presents general characteristics of commonly available optical satellite imagery. Next, with the development of sensors, this paper suggests that data fusion of SPOT-5 and Sentinel-2 may be helpful in LCLU change detection caused by refugees/IDPs or conflicts. Smallsat imagery may be promising for research on humanitarian applications that require a high temporal frequency of imagery.
Multi-feature sample database for enhancing deep learning tasks in operational humanitarian applications
Stefan Lang1, Lorenz Wendt1, Dirk Tiede1, Yunya Gao1, Vanessa Streifeneder1, Hira Zafar1, Adebowale Daniel Adebayo1, Gina Maricela Schwendemann2, Peter Jeremias1
1Christian Doppler Laboratory for Geospatial and EO- Based Humanitarian Technologies, Department of Geoinformatics – Z_GIS, University of Salzburg, Salzburg, Austria; 2Spatial Services GmbH, Salzburg, Austria
Amongst the many benefits of remote sensing techniques in disaster- or conflict-related applications, timeliness and objectivity may be the most critcal assets. Recently, increasing sensor quality and data availability have shifted the attention more towards the information extraction process itself. With promising results obtained by deep learning (DL), DL is not agnostic to input errors, in particular in sample-scarce situations. The present work seeks to better understand the influence of different sample quality aspects propagating through convolution network layers in image analysis. We systematically assess the quality of input samples under the following aspects, recorded next to each sample’s label (category, class): (1) herited properites (quality parameters of the underlying image such as cloud cover, seasonality, etc.); (2) individual (i.e, per-sample) properties, including a. lineage and provenance, b. geometric properties (size, orientation, shape), c. spectral features (standardized colour code); (3) context-related properties (arrangement). In an initial stage, several hundreds of samples collected from different camp settings were selected and annotated with computed features. This annotation is automatized in a way that many thousands of existing samples can be labeled with this extended feature set.
Testing transferability of deep learning-based dwelling extraction in refugee camps
Getachew Workineh Gella1, Lorenz Wendt1, Stefan Lang1, Andy Braun1, Dirk Tiede1, Barbara Hofer1, Yunya Gao1, Barbara Riedler1, Ahmad Alobaidi2, Schwendemann Gina Maricela2
1Paris Lodron University of Salzburg (PLUS), Christian Doppler Laboratory for Geospatial and Earth Observation based Humanitarian Technologies (GEOHUM), Austria; 2Spatial Services, Austria
For effective humanitarian response in refugee camps, reliable information concerning dwelling type, extent, surrounding infrastructure, and respective population size is essential. As refugee camps are inherently dynamic in nature, continuous updating and frequent monitoring is time and resource-demanding, so that automatic information extraction strategies are very useful. In this ongoing research, we used labelled data and high-resolution Worldview imagery and first trained a Convolutional Neural Network-based U-net model architecture. We first trained and tested the model from scratch for Al Hol camp in Syria. We then tested the transferability of the model by testing its performance in an image of a refugee camp situated in Cameroon. We were using patch size 32, at the Syrian test site, a Mean Area Intersection Over Union (MIoU) of 0.78 and F-1 score of 0.96, while in the transfer site, MIoU of 0.69 and an F-1 score of 0.98 were achieved. Furthermore, the effect of patch size and the combination of samples from test and transfer sites are investigated.
Extraction of dwellings of displaced persons from VHR radar imagery – current challenges and future perspectives
Department of Geoinformatics, University of Salzburg, Austria
While many studies exist to identify buildings from optical satellite images, radar-based approaches are still missing in humanitarian contexts. The reasons are delayed technological development of satellites, challenges related to scattering mechanisms returning from huts, tents, informal dwellings, and their natural surroundings, but also geometric distortions caused by the side-looking radar aperture. These challenges can be overcome by image enhancement or multi-image composites, but also by advanced methods on building extraction, such as convolutional neural networks (CNNs).