Forest monitoring using remotely sensed data is a central task in forestry and ecosystem studies. The long-term assessment of woody vegetation assists, for instance, in change detection and handling environmental hazards. Recently, France has made available its historical aerial images archive, covering the whole country extent. The public availability of such datasets expands the scope of long-term monitoring studies, such as forest monitoring. The main aim of this study is to investigate how the volume and the quality of the reference data influence the semantic segmentation of woody vegetation using historical grayscale aerial images in southwestern France. The two main contributions of this study are the provision of a rare 1942 binary dataset (woody vegetation vs. no woody vegetation) and the investigation of the effects of training data quality and quantity using a deep learning architecture.

Detecting building changes using earth observation datasets has numerous applications, particularly in humanitarian emergency responses. Deep learning models for change detection have shown considerable progress, but they require large annotated datasets for training and validation. Although semi-supervised change detection methods utilize a large amount of unlabeled data from the same domain, their effectiveness is limited when applied across different datasets due to shifts in data distribution. We proposed a cross-domain semi-supervised change detection approach that incorporates consistency regularization and a deep change domain mapping strategy to handle data distribution shifts. This method achieved a change class Intersection over Union (IoU) of 0.57 and 0.22 in the challenging cross-domain scenarios of LEVIR-CD → WUHAN-CD and LEVIR-CD → EGY-BCD open datasets, respectively. Additionally, we tested this approach for monitoring building changes in temporary settlements hosting forcibly displaced people (FDP), using cross-domain data from open datasets and images of these temporary settlements. Despite the improvements introduced by domain mapping, the results are not yet sufficient for practical use in emergency response operations.

In supervised learning, deep learning models demand a large corpus of annotated data for object detection and classification tasks. This constrains their utility in humanitarian emergency response. To overcome this problem, we have proposed an unsupervised dwelling counting from very high-resolution satellite imagery by combining a Variational Autoencoder(VAE) with an anomaly detection approach. When VAEs are applied in earth observation for dwelling localization and counting, we observed two critical limitations (1) the balance between reconstruction and good latent code, where in-favour of good reconstruction of dwellings leads to weak anomaly score maps that fail to properly localize dwellings (2) limited spatiotemporal invariance of the learned latent code. When the model is trained with datasets obtained from different geography and time, it fails to properly localize dwellings. For the first problem, we introduced self-supervision by creating synthetic anomalies. For the second problem, we introduced latent space conditioning. The approach is tested on 9 very high-resolution images obtained from six Forcibly Displaced People settlement areas. Results indicate that combining VAE with an anomaly detection approach has reached an AUC value ranging from 0.70 at complex settlements towards 0.98 at relatively less complex settlement areas. Similarly, an MAE value of 56.67 towards 5.03 is achieved for dwelling counting. Joint training of combined datasets with latent space conditioning and self-supervision enabled the achievement of results better than classical VAE, with improved spatiotemporal transferability of the model with more crisp and strong anomaly maps. Overall implementation code will be available at https://github.com/getch-geohum/SSL-VAE.

This work studies change detection from satellite image time series (SITS) with a proposed framework that leverages SITS using self-supervised learning. Experimental evaluation conducted on a study area in southwestern France demonstrates the effectiveness of the approach, with varying quantities of labeled training data. The results highlight the potential of self-supervised learning in producing accurate change detection maps for land cover analysis.

Even though computer vision models are excellent for automatic scene segmentation and object identification from remotely sensed imagery, they demand a huge corpus of annotated data for the training and validation which is a huge challenge in humanitarian emergency response. To tackle this problem, we propose unsupervised dwelling object counting combining Variational Autoencoder (VAE) with an anomaly detection approach. The approach is tested in six Forcibly Displaced People (FDP) settlement areas situated in different parts of the world. Using an anomaly map computed with the VAE model, we demonstrated the possibility of properly locating dwelling objects using anomaly maps. Dwelling counts are obtained by further segmenting anomaly maps. Results show that, though it has strong spatio-temporal variation, the VAE model exhibits promising potential for locating and counting dwellings. It is also observed that in FDP settlements with dense buildings and extremely low contrast between buildings and ground or environment, the performance is relatively lower than the performance achieved in settlement areas with regularly spaced and less complex building structures.

The continuously increasing amount of Synthetic Aperture Radar (SAR) data is becoming an important complementary source of information for humanitarian action especially when the availability of optical remote sensing is limited. The SAR-inherent speckle effect remains an obstruction to SAR image interpretation and automated analysis. Recently, deep learning-based despeckling techniques such as supervised non-local methods have shown excellent speckle reduction for very high resolution (VHR) SAR data. Because the approaches have been mostly applied to VHR commercial SAR datasets, this work transfers, adapts and tests an existing deep learning-based non-local speckle filter on medium-resolution Sentinel-1 data. The results show that adaptations to the model lead to superior filtering results compared to conventional filters while simultaneously yielding better performance when the filtered data is used in a downstream task relevant to humanitarian application - dwelling change detection in forcibly displaced population settlement areas.

Riparian forest habitats are biodiversity hotspots providing crucial ecological services, where Earth observation information can be used to analyze prevailing habitats, assess their quality and conservation status, as well as the change dynamics over time. Here we present advancing investigations in the Salzach river floodplain (`Salzachauen') at the Austrian/German border, using multiannual/multiseasonal VHR data coverage. Object-based image analysis is applied to delineate forest habitats on different hierarchical levels. To enhance the stability of the basic segmentation levels, the usage of multitemporal WorldView-2 data is investigated. Based on five panchromatic bands, we created (1) temporal stable and textural homogeneous core segments and (2) edge segments caused by sensor driven data shifts and other effects.

Riparian forests are well known for their high biodiversity and critical ecological services. Earth observation-derived information can be used to analyze these habitats and their quality and conservation status, as well as change dynamics over time. In this study, object-based image analysis methods were applied to (semi-)automatically delineate forest habitats and to assess habitat quality in the Salzach river floodplain. Multitemporal and multiseasonal satellite imagery, in particular, WorldView-2 images from 2011/2012 as well as a SPOT-5 scene from 2005, was used. To perform validation on different levels of thematic resolution, locations of single trees of dominating species were collected, and tree species dominance and vegetation density were documented. The main pillar of the described workflow is the use of an object feature library (OFL). Conceptually, the OFL is needed to calibrate the method of information extraction and habitat quality estimations based on satellite imagery of different temporal, spectral, and spatial resolutions. This is considered a critical step to increase the robustness and transferability of the rule-based semantic classification approach.